Britain is staking its economic future on artificial intelligence (AI) hardware, committing £1.1 billion ($1.4 billion) to build a sovereign chip industry, a national supercomputer, and a generation of engineers capable of competing with the world's largest technology powers.

The plan spans a £750 million national AI supercomputer, a £120 million innovation program to fund the design and testing of novel chips, and a new investment fund backed by up to £150 million from the British Business Bank. The government is also committing £80 million in total to skills.

“On AI, Britain has three options. We could stick our heads in the sand and hope for the best. Or you can remove the guardrails completely and ignore the consequences. Or you can take a third path, where we back the British businesses creating the jobs and technologies of the future, but never lose sight of who that progress must serve,” said Keir Starmer, Prime Minister of the United Kingdom.

“What matters is whether Britain is ready for it. Whether British innovators have the tools, the investment and the backing they need to lead it. And whether working people have the opportunity to benefit from it,” he said.

Starmer said the third path meant British tech companies starting, scaling and staying in Britain, with the rewards felt across the country. He said government must be active in technology policy, supporting risk-takers and providing the conditions for businesses to thrive, while also ensuring national sovereignty and giving working people security through change.

He said innovation was in Britain’s DNA. The country had invented the World Wide Web, pioneered the jet engine and powered the Industrial Revolution. It was the job of government not just to recognize the scale of that talent but to match it with the scale of its ambition.

He said ideas being born in Britain were not enough. They had to be able to grow there too, which was why the government was crowding in capital, betting on British businesses and creating confidence for investors worldwide.

The global AI chips market is forecast to reach one trillion dollars in the early 2030s. Securing just 5% of that market would generate £50 billion in revenue and create tens of thousands of highly paid jobs in the UK.

Of the £750 million earmarked for the national AI supercomputer, £400 million will go toward next-generation chips. £150 million of that will buy next-generation inference chips this summer, creating an immediate opportunity for British firms. A further £250 million will support more specialized chips as the most promising technologies mature.

Light over electricity

Starmer made his remarks at London Tech Week 2026, the annual technology conference held in London and organized by London & Partners, on June 8. A government press release published the same day set out the full details of the AI Hardware Plan, covering chip procurement, supercomputing infrastructure and skills investment.

Among the most striking announcements was the expansion of the Scaling Inference Lab, delivered by the Advanced Research and Invention Agency (ARIA) and CommonAI, into a national capability for testing and validating AI chip technologies. At least £20 million from the AI Hardware Innovation Program will fund the expansion.

The Lab is already delivering results. British AI company Oriole Networks, working with AMD through the Lab, will deploy the world’s first large-scale AI system that uses light rather than electrical signals to move data between chips. The photonic interconnect technology is designed to significantly boost the performance of UK data centers.

The shift from general-purpose chips to bespoke hardware plays directly to the UK’s strengths. British companies are already leading the next generation of AI hardware. Arm’s chip designs are used in everything from smartphones to AI data centers. Startups Fractile and Olix have raised more than £320 million between them.

A £120 million AI Hardware Innovation Program will provide companies with funding to design, develop, and test novel chips before they know whether the underlying technology will succeed commercially. The government described this as a way for Britain to ensure the next generation of world-leading chip companies is grown at home.

“Last year, I set us a target of upskilling seven and a half million workers with AI training by 2030. Today, I am pleased to report back that 1.7 million workers have already received that training,” Starmer said.

He said these figures came one year after he had set the target at the same stage. He added that while the government could not know exactly where AI would cause disruption, it knew how it would respond and whose side it would be on.

The skills investment covers a new £12 million Centre for Doctoral Training in Chip Design and a £20 million expansion of the TechFirst program to support 500 more Doctor of Philosophy (PhD) students. Undergraduate semiconductor bursaries will rise to 500 per year. Arm has joined as a strategic partner for TechFirst.

He said the government was rolling out AI tutors to 450,000 children on free school meals to close the attainment gap, and announced a new AI jobs tool to help people out of work find employment, create their curriculum vitae (CV) and return to the workforce.

Warrington and beyond

The story Starmer chose to open his speech was not a statistic or a policy announcement. It was a factory in Warrington.

“For centuries, Warrington was at the forefront of Britain’s soap-making industry. And until a few years ago, an Unilever factory was its epicenter. Generations of local people worked there. Families built their lives around it. But then it closed. And for many people in the town, that factory became a symbol of a community left behind,” he said.

That factory was now being transformed into a new AI data center, Starmer said, bringing skilled jobs and investment to a town that had watched opportunity pass it by. He said similar stories were unfolding in Lanarkshire, Liverpool and Leeds.

“Young people can look at that site and see not what their community used to be, but what it can become,” he said.

AI was already delivering faster diagnoses in the National Health Service (NHS), reducing court backlogs and speeding up planning decisions. He described these as real benefits that people across Britain are feeling right now.

UK startups had raised close to half of all European tech investment in 2025, a figure Starmer described as a profound achievement. Britain was the third-largest technology economy in the world. Each investment, he said, was an endorsement of British talent, British industry and the approach the country was taking.

He cited Reflection AI as expanding in Britain, creating 1,000 roles over three years because of the country’s talent pool. AMD was among the companies doubling down on the UK. The government had also created a Global Talent Taskforce to strengthen the routes bringing exceptional innovators to Britain.

Liz Kendall, Secretary of State for Science, Innovation and Technology, who announced the AI Hardware Plan at the same event, said the UK was already a global leader in chip design and believed it was a race Britain could win.

On children’s safety, Starmer said the pace of technological change could not be an excuse for harm. He called on tech companies to introduce device controls preventing children from sending and receiving sexually explicit images, citing an earlier incident involving the Grok platform.

“If they choose not to, then we will act. And we will change the law. When it comes to the safety of our children, standing by is not an option,” he said.

He said the question was not whether the AI revolution was coming, but whether Britain would shape it or be shaped by it. He said the government had made its choice: to take control, to be ambitious about what Britain could achieve, and to make AI work for every part of the country.

Artificial intelligence (AI) companies consistently disclose only what benefits them commercially, leaving consumers, investors and regulators without the information they need to make meaningful decisions about the systems they use.

Open Ethics is a non-profit initiative with a mission to make AI systems more accountable and understandable. Co-founded by Alice Pavaloiu and Nikita Lukianets, it builds open-source tools that allow product owners to describe their AI solutions in a standardized, accessible way.

“Most organizations who fill out these disclosures are going to use them in two ways,” Pavaloiu told TechJournal.uk in an interview in London. “The first is to create a public disclosure to become more accountable, align with compliance frameworks and strengthen their reputation with investors. The second, which is an indirect benefit, is to start having this conversation about transparency internally and create a database of incidents.”

“The end user wants to know whether the system hallucinates or whether their data is safe. An auditor or software developer wants architectural detail. An investor looks at totally different transparency metrics,” she said.

Most organizations have never formally audited their own AI failures. The disclosure process, even before anything is made public, forces that conversation to happen.

Because labels are self-disclosures, organizations can in principle submit inaccurate information. Open Ethics is working to address this by partnering with third-party auditors to cross-check published disclosures against the systems they describe.

The initiative’s portfolio spans four disclosure tools: the Open Ethics Label, a standardized transparency label for AI products; the Open Ethics Data Passport (OEDP), which maps dataset construction and tests for bias; the Open Ethics Maturity Model (OEMM), a five-level ethical governance framework; and the Public Surveillance Transparency Project, which applies the same principles to cameras in public spaces.

Label as mirror

Pavaloiu spoke about the suite of disclosure products Open Ethics has developed and the obstacles that remain.

She and Lukianets came to the transparency problem after identifying persistent gaps in communication between AI providers and the public, and between the technology industry and government.

When they began engaging with organizations around five or six years ago, they found both sides operating in silos: governments unable to form policy without industry input, and companies resistant to openness for fear of commercial exposure.

Their response was to look at how other industries handle consumer information. They visited supermarkets and household appliance stores to study product labels and research which information genuinely helps a consumer decide whether to use a product.

“We were thinking about these very long privacy policies, obviously written in legal language, quite dry and very hard to understand. We said, what about labels? They are quite compact and offer a transparent view inside the product,” Pavaloiu said.

“We realized that technology should be regulated equally to food, but it does not come with a label. So we said, why would we not create one.”

The Open Ethics Label is generated through a structured form that covers three pillars: training data, algorithms and the decision space. An optional fourth section covers data processing, retention and human-in-the-loop practices.

Once submitted, Open Ethics adds a timestamp, issues a cryptographic SHA3-512 signature to ensure data integrity, generates a machine-readable JSON file and returns embeddable HTML for the organization’s website.

“Designing the label was not actually hard. Having it adopted by organizations in a voluntary way: this is where our challenges began,” she said.

The label is not an ethical certificate, and Open Ethics is deliberate about that. Whatever an organization inputs, the protocol outputs in a standardized, readable form.

“We always say the Open Ethics Label is not a certificate. It is rather just a mirror. Whatever they input, we output in a standardized way. This is all the magic that goes behind it,” Pavaloiu said.

The Open Ethics Data Passport goes deeper into the structure of AI systems. Where the label is an invitation to transparency, the passport is a detailed snapshot of how a training dataset was constructed and how a system was built.

Pavaloiu said the data passport was designed to deliver built-in transparency and allow organizations to spot systemic bias in trained AI models across all phases of the machine learning lifecycle.

“This is where it gets its strength from,” she said. “Think about a self-driving car with many components and sensors. If every vendor of every AI component generates a data passport at the point of assembly or sale, the car can come with a compounded data passport of every single component. Whenever something goes wrong, you have an entire map: a starting point for testing for bias.”

The human bias problem

Technical bias in AI systems has attracted significant attention and a growing set of tools. Pavaloiu argues that human bias, embedded during the data labeling process, is a far larger and harder problem that the industry has barely begun to address.

“We are addressing technical bias, and there are quite a lot of good tools on the market that do that. But what about human bias? This is the far bigger and harder problem to address,” she said.

Open Ethics is developing a hypothesis: if a project owner identifies values relevant to their application, for example dignity in a medical AI context, and hires data labelers who already hold those values, the resulting dataset should carry less human bias.

The challenge is designing a value assessment methodology that cannot be gamed by respondents.

“If we just ask them: are you trustworthy? Everyone will say yes to every value. Harvard has an implicit association test, essentially a speed test where the person answers based on whatever their mind associates in a split second,” Pavaloiu said. “It bypasses the conscious and takes you straight into the subconscious. This could be a way to identify values rather than simply asking.”

Open Ethics has been self-funded by Pavaloiu and Lukianets since the initiative launched. A membership scheme introduced around a year ago set fees at 50 euros per year, designed to cover basic running costs. A free contributor tier ensures students and early-career practitioners can participate.

“We realized we would probably need more than two minds to scale and drive impact across the globe. We have welcomed 100 members and contributors: AI experts, regulators, legislators, AI enthusiasts and people from law, philosophy and the medical field,” she said.

The most contentious project in the portfolio takes the disclosure model into physical space. The Public Surveillance Transparency Project proposes that institutions operating cameras in public locations post a printed label beside each camera with a QR code linking to a full disclosure.

Pavaloiu said the idea grew partly from personal experience. When she lived in Estonia, delivery robots roamed city streets, drawing smiles from pedestrians and stopping obediently at crossings. But the machines were fitted with 360-degree cameras, and no information was posted anywhere about who operated them, what data they collected or where it went.

“We never thought privacy and safety need to be opposing forces. There is a very big imbalance in power and information between the person who is surveilling and the person being surveilled,” she said. “Usually we do not have much indication. We do not know what type of surveillance it is, whether it is biometric, who the beneficiary of the data is, how long it is stored, or whether it is sold to third parties.”

The project aims to answer those questions at the point of surveillance itself. Under the proposal, anyone passing a camera in a public library, bank ATM or any other institution could scan a QR code on a posted label and immediately access a full disclosure of how their data is being handled.

The project has faced the strongest resistance of any Open Ethics initiative. Pavaloiu said she hoped stronger regulatory requirements, including provisions in the EU AI Act, would create the conditions that voluntary adoption has so far failed to deliver.

Open Ethics remains focused on advancing transparency standards across the industry. Its ultimate aim is a digital space where AI systems can clearly explain themselves to the people who use them.

The grinding analytical work that fills the days of lawyers, bankers and management consultants is disappearing, not gradually, but at a pace that most enterprises have neither anticipated nor prepared for.

Artificial intelligence (AI) is eliminating the middle layer of knowledge work, and the professionals who have built careers on it are running out of time to adapt.

The shift is more disruptive than previous waves of automation because it strikes at cognitive rather than manual labor. Unlike factory automation or robotic process automation, which displaced workers at the lower end of the skills spectrum, the current wave is arriving from the top.

“It is pretty naive to say that transition of work is not happening. There is work that used to require a human to be involved which is not the case anymore,” Christian Lund, co-founder of Templafy, told TechJournal.uk in an interview.

“What remains super important is two things: the direction setting, meaning how to ask the question and what you want to achieve, and the ability to assess the quality of the output,” he said. “Everything in the middle, all the analysis and putting together decks and documents, that is what is going away.”

“All the highly paid, sought-after, highly educated people, including lawyers, bankers, accountants and consultants, will need to reposition themselves to play a new role, with AI being stronger and stronger. It is very hard to compete with because AI has all the brainpower imaginable. You have to be able to use it rather than produce it,” he said.

He added that the shift had also overturned a long-held assumption about the nature of questions. The old saying that there are no stupid questions, only stupid answers, was in his view now completely wrong. With AI, the quality of output depends entirely on the quality of input, making precise prompts the most critical human skill in knowledge work.

Templafy was founded in Copenhagen in 2014 and is now headquartered in the United States. It helps large organizations automate the generation of business documents, presentations, emails, spreadsheets and contracts, serving more than four million knowledge workers including clients at Adobe and across the Big Four accounting firms, major banks and leading law firms.

The company’s customer base spans North America and Europe in roughly equal measure, concentrated in industries with high proportions of knowledge workers: investment banks, hedge funds, accounting firms, law firms and large consultancies. Templafy has around 150 employees, of whom approximately 45% are engineers.

Control lost to AI

The interview covered enterprise AI adoption and the growing gap between what AI promises and what it delivers at scale.

The displacement of knowledge workers sits alongside a parallel crisis inside the organizations that employ them. Enterprises have lost control of how AI is being used, and the consequences, including compliance failures, inconsistent outputs and reputational risk, are becoming impossible to ignore. A PwC survey found 56% of companies are seeing no return on their AI investments.

“The main concern about adopting AI in general is that things go rogue, that they have no control,” Lund said. “They used to be able to apply control by putting a piece of software on people’s computers and saying: this is what we want you to do. ChatGPT changed that. It has been democratized, which in an enterprise context is pretty disturbing.”

He said enterprises had gone through a rapid sequence of responses: initial excitement at what AI could do, followed by alarm at losing control, then a more pragmatic search for tools that could deliver productivity gains without exposing the business to risk. The old model of controlling employee behavior through prescribed software workflows was gone.

“Enterprises want to make sure that as users get stuff done, they inject themselves in the middle to make sure it is done in a way the business can live with, that does not put liability and risk on them. More productivity and efficiency without the risk is what they are looking at,” he said.

Against that backdrop, Lund identified three areas where Templafy distinguished itself from general-purpose AI tools and large language model (LLM) providers: quality, consistency, and cost. On quality, he said general AI tools could produce visually impressive outputs but failed at the specificity required for enterprise documents.

“We have been using the term: is it a toy or is it a tool? It is impressive what can be done if you provide a prompt to an LLM, but it is hard to use for much. If you are doing a proposal, for example, you cannot have margins of error. It has to be accurate every time,” he said.

On consistency, free-form prompting inevitably produced divergent results across a workforce. Ten users trying to produce similar documents would ask slightly different questions of potentially different models and get very different outputs. On cost, he described a model-routing approach that assigned different AI models to different parts of a workflow.

“Let’s have Claude do the brain work, where it is really difficult: the research, the analysis. But we might use a different model to actually produce the deck, because it is a lot cheaper,” he said. “We are distributing the workload the way you would in a non-AI world, where you use different skill sets and different pay grades for different tasks.”

Patenting the instruction layer

To address the control problem at its root, Templafy has developed a patented approach to generating what Lund called “instruction books” for AI.

Rather than relying on individual users to write effective prompts, the system captures a user’s intent and dynamically generates a detailed, contextually rich instruction set passed to the AI model before it begins work.

“If you want to get to highly reliable results when you build business documents with AI, you have to be very instructive,” Lund said. “We have a patented way to build instruction books for AI. We can build these dynamically on the fly, so instead of eight words from a user, it would be extrapolated into 30 pages of very specific context, including data, so the AI can do a much better job.”

The technology draws on Templafy’s decade-long history in document template automation. The company has also built Model Context Protocol (MCP) integrations, allowing it to operate as an embedded capability within Claude, Gemini, Microsoft Copilot, and other leading AI environments, so users need not switch platforms to access its document-generation tools.

“We prefer to support users working where they already work. Businesses are picking one or two of the bigger LLMs and saying: that is our starting place for work. ” We want to make sure we are just like an ingredient or an enzyme that lives inside that,” he said.

Lund said Templafy was already partnering with major LLM providers including Microsoft and several newer entrants, and expected those relationships to deepen significantly. He said major LLM providers were increasingly seeking specialist partners to bridge the gap between general-purpose AI capability and enterprise-grade use cases.

“We are not trying to compete with the LLMs. What we want to do is make Claude, OpenAI, Gemini and Copilot better than they are today by filling in the gaps,” he said. “If you go bottom-up and have a very product-led approach, you will get a lot of users very quickly, but it is difficult to sustain your business. If you want to make a good business on AI, you have to get closer to the actual use cases.”

Artificial intelligence (AI) is poised to compress the automotive industry’s most time-consuming work, from the first sketch of a product concept to its appearance in a finished vehicle. The pressure to move faster is intensifying across every segment of the market.

Leading carmakers say the gains are most visible inside the organization rather than on the showroom floor. AI is accelerating the validation of ideas, the engineering of components and the deployment of automation on shop floors, a transformation that will ripple from product design all the way to the customer. The shift is already underway at some of the world’s largest vehicle manufacturers.

“AI will speed up the process from the very first idea to seeing it implemented in the car, all the way from deriving a business case, validating a product with the customer, engineering, simulating, crash simulation, to physical AI on the shop floor. We will definitely see speeds in this one,” said Manuel Schneider, head of open innovation at BMW.

Schneider said BMW has been working on AI for 15 years. His open innovation role involves identifying and validating ideas from academia, startups and entrepreneurs worldwide, as well as from more than 120,000 employees inside the company. He described AI adoption as a cultural shift rather than a departmental initiative, one that requires the entire organization to become receptive to external ideas.

“The organization needs to open up. We need to open up to large language models (LLMs), we need to open up to Chinese LLMs and bring world knowledge into the company,” he said.

Karin Svensson, chief sustainability officer of Volvo Group, one of the world’s largest manufacturers of commercial vehicles, trucks, buses and construction equipment, said AI is central to creating business value across its entire value chain, from design and production to customer interfaces, operation and services. She said deployment is uneven: some areas are quite advanced, while others are still at a sorting stage.

“AI could absolutely be something that can increase sustainability. But we cannot only talk about AI for sustainability. We also need to talk about sustainable AI,” she said.

China sets the pace

The panel “Shifting Gears with AI: The Future of Mobility” was moderated by Vijay Vaitheeswaran, global energy and climate innovation editor of The Economist, at the 12th annual Sustainability Week organized by Economist Enterprise in London.

Schneider said BMW takes a technology-agnostic approach, working with multiple LLM providers, including Gemini, Anthropic, OpenAI and DeepSeek, depending on the use case. He has team members based in China, Germany and the US who exchange frequently on AI developments across all three ecosystems, giving the company a ground-level view of how different AI cultures are evolving.

Bono Ge, country manager for the UK and Ireland at BYD, said the difference in customer behavior between China and Europe is stark.

“The customer is quite different in China. Chinese customers enjoy electric cars because the infrastructure is already there, and as a further extension, there is a lot of smart stuff they are more willing to explore,” Ge said.

He said European customers surveyed at random had often never used the voice control on their BYD vehicles. Chinese customers, by contrast, actively seek out and test every digital feature available to them. He said he expects European consumers to follow a similar path once they become more comfortable with electrification.

“We see massive speed in China, in the market in general, across various companies. To us, it’s super motivating. When I see my colleagues and competitors using local AI and local technology and advancing, this is something we have to do. There is no way around it,” Schneider said.

BMW is launching the iX3 in China in partnership with a local company to push autonomous driving capabilities, with the vehicle expected to reach the market in autumn 2025. Avoiding vendor lock-in is a guiding principle for the company, applying equally to vehicle drivetrains and to AI service providers.

“We have to use the ecosystems that are provided in the markets where we operate. We see many advancements in China that are highly beneficial to the world. It would be wrong not to look into them and use them,” he said.

The shift in innovation culture extends beyond software. Schneider said BMW is piloting in-car medical services in partnership with a hospital in Berlin, exploring whether the vehicle’s sensors can monitor passengers’ heart rate variability and blood pressure. The pilot represents a broader opening of what has historically been a tightly closed innovation ecosystem.

“We have a lot of sensors in the car and we have the people in the car. Can we measure heart rate variability, blood pressure? Can we provide medical services? This is at a pilot stage, but it just shows that this very closed ecosystem is opening up to new transactions and new business models,” he said.

Autonomous driving at levels 3 and potentially level 4 will create unoccupied time inside vehicles, he said, and in-car medical monitoring is one potential use of that time.

Autonomous driving is classified on a scale from level 0 to level 5:

• 0: the driver controls everything

• 1 and 2: basic driver assistance such as lane keeping and adaptive cruise control; the human remains in charge

• 3: the car handles most driving tasks, though the driver must be ready to take over

• 4: full self-driving in defined conditions with no human intervention needed

• 5: complete autonomy in all conditions and on any road

Beyond borders and batteries

Ge described the automotive industry’s evolution using a football analogy. The first half is electrification, which is nearing completion in several markets: Norway has exceeded 95% electric-vehicle (EV) penetration in new-car registrations, and China has reached 50%. He said the second half belongs to AI-driven digitalization, and the race to lead it is already underway.

“The car must be smart enough to know which things you want the car to do, and which things you are saying to the passengers. The charging route can also be fine-tuned by AI,” he said.

AI applications in development at BYD include autonomous driving for taxis and buses, natural language interfaces, battery health monitoring and vehicle-to-vehicle messaging. He said battery state-of-health monitoring, which tracks degradation over time to predict remaining useful life, is already among the most mature applications in the fleet. Regulatory approval will be the key gating factor for many others.

Svensson raised the resource cost of AI itself. The semiconductor chips required for advanced AI applications consume significant amounts of fresh water during production, a cost that is rarely factored into assessments of AI’s environmental impact.

She said companies face a transparency challenge in marketing AI systems as “smart” while being honest about their limitations and failure rates. The transportation sector must make mobility cleaner, safer and more efficient as global demand continues to rise. AI can contribute to that goal, but only if its own environmental footprint is managed alongside the benefits it delivers.

Trade tensions cast a long shadow over the panel. Ge said China, South Korea and Japan together dominate global power battery production, with China alone supplying 70% of EV batteries worldwide.

“If we keep talking tariffs but cannot develop electric cars because there is not much battery production in Europe, the whole industry suffers. The automotive industry is global. We have to ensure global sustainability for the supply chain,” he said.

Schneider agreed that the codependencies run too deep to unwind through trade barriers alone.

“There is no car that doesn’t have ingredients and parts from China, Germany and the US. The value chain is established and there are codependencies. We have to account for these tariffs, and at the same time believe in the dependency between China, Europe and the US,” he said.

The codependency he described has already shaped policy.

In January 2026, the European Union withdrew the punitive anti-subsidy import tariffs it had levied on Chinese electric vehicles, replacing them with a minimum selling price agreement.

As AI capabilities mature and the EV transition accelerates, both the opportunity and the pressure on the global automotive supply chain will only intensify.

The energy transition has a paradox at its heart. The technology most needed to accelerate it is also one of its fastest-growing consumers of power. Artificial intelligence (AI) is driving electricity demand to levels that grids struggle to meet, yet it is also emerging as one of the most powerful tools for cutting the energy waste that has frustrated efficiency targets for decades.

The case for urgency is stark. Despite offering some of the fastest investment paybacks available, energy efficiency has been chronically deprioritized by companies and governments alike. The world has consistently missed its own targets, including those set at COP28.

“What we need is energy intelligence. The first thing when we are in a company's boardroom is: why are you not implementing energy efficiency? The payback is one to three years depending on the energy conservation measures (ECMs) you have, and it’s the best return on investment,” said Raquel Espada, vice-president of energy services and sustainability at SE Advisory Services, a division of Schneider Electric.

“Energy efficiency has been the hidden gem, and nobody has been thinking about it. It is critical because we are talking about sustainability and return on investment (ROI). The best energy is the one that is not produced,” she said.

Schneider Electric supports companies globally on decarbonization roadmaps through its advisory services arm. Espada said AI is now changing the calculation by making invisible waste visible.

Building management systems powered by machine learning can identify exactly where energy is being lost, removing one of the most persistent barriers to action. Paybacks of one to three years are available to companies willing to look.

The data center industry is among those now demonstrating what AI-driven efficiency can deliver in practice.

Doug Adams, president and chief executive of NTT Global Data Centers, the third largest data center platform in the world with 150 facilities across 21 countries, said his company has deployed AI across its sensor networks.

“We deployed AI in our data centers so that we can monitor sensors in real time and drive 20% of our power cost out by monitoring the sensors and reducing fan pressure,” he said.

He said operating a data center involves thousands of sensors that generate what he described as “alarm overload,” a volume of alerts no human team can process. AI-driven correlation cuts through that noise to identify where consumption can be reduced.

His company also uses direct liquid-cooled chips across hundreds of megawatts of capacity. This reduces power draw by around 5% compared with traditional air cooling, equivalent to five megawatts saved across a single 100-megawatt facility.

NTT channels rejected heat from its German data centers to warm 10,000 homes. Adams said the arrangement improves the company’s bottom line as much as its carbon footprint.

Why efficiency keeps losing

The discussion took place at the 12th annual Sustainability Week, organized by Economist Enterprise and held in London. It was moderated by Vijay Vaitheeswaran, global energy and climate innovation editor of The Economist.

Vaitheeswaran opened by noting that despite fast paybacks and clear returns, governments and companies have consistently failed to meet their own efficiency targets, including commitments made at COP28.

Espada said the barriers have historically included financing constraints, short-term payback expectations and a lack of data.

“For me, sustainability is a financial value creator, and efficiency is going to be the first topic on the table because of cost reduction. Sometimes people didn’t have the data, and that’s where artificial intelligence is supporting efficiency right now, because you can really get the data on where you are wasting energy,” she said.

Sophie Graham, chief sustainability officer of IFS, an industrial software and AI company, said her firm’s experience across large field-service organizations provides real-world evidence of what AI-driven efficiency looks like at scale.

“IFS works with the 70% of the workforce that’s not behind the desk, but is in the field, on the factory floor. When we talk about AI, we’re talking about embedding it within those industrial processes. It is not the entertainment ChatGPT. It is really built into the processes and systems that ultimately run the country,” she said.

“We use AI-driven scheduling and optimization over a network of engineers, and we found, in a study last year with over 60 customers, that on average it reduces travel distance by 37%. That’s a huge reduction if you scale across tens of thousands of engineers and many different companies,” Graham said.

The gain simultaneously cuts carbon, raises productivity and reduces costs.

Power grids and buildout

One of the most practical applications discussed was the use of AI sensors to unlock transmission line capacity without new infrastructure. Graham described a customer deploying dynamic line rating, with AI-driven sensors on transmission lines that can unlock around 30% more capacity on existing lines.

“Without building out more, we can get more capacity on the grid, which in turn allows us to bring in more renewables,” she said.

Line ratings have historically been based on predicted rather than actual weather conditions. Real-time data frequently shows lines can safely carry far more power than static ratings suggest. With “not in my back yard” (NIMBY) opposition blocking new transmission projects across Europe and the US, extracting more from existing infrastructure has become urgent.

She also pointed to Ireland as an early regulatory test case. Data centers there consume 22% of the country’s electricity. A new mandate requires new facilities to source 80% of their annual energy from new clean energy projects.

“Microsoft is unveiling its data centers there. It’s one that other energy-constrained countries such as the Netherlands and the US are closely watching,” she said.

Adams said the scale of the global infrastructure challenge is difficult to overstate.

“Only in the last two years has anyone even known what a data center is. I spent the first 23 years explaining to people that the data center was a hotel for computers. Now it’s on the tip of everybody’s tongue, and I’ve seen more change in the last two or three years than in the previous 20-plus years,” he said.

A McKinsey report he cited estimated between three and seven trillion dollars of new investment will be needed just to meet data center demand through 2030. Bridging solutions involving natural gas turbines are being deployed as a temporary fix while grid capacity catches up.

He pushed back against media distortion of the sector’s energy footprint.

“AI is a tremendous technology when applied responsibly, and the key is responsibly. Data centers account for about 1.5% of the power globally. They don’t use a third of the power like the media sometimes makes you believe,” he said.

Espada said the question has moved rapidly up the boardroom agenda.

“This is a question now on boards. CEOs are asking not why I’m going to use AI, but how I’m going to use it. What is the environmental impact? It’s not only energy. It’s water, materials, resources, and sometimes the community impact,” she said.

She drew a parallel with renewable energy projects that once faced community resistance in rural areas, and said data centers are now encountering the same dynamic.

Electrification of industrial processes is emerging as a key resilience strategy for companies seeking to reduce gas dependency. Espada cited Roca, a Spanish ceramics manufacturer, as a case study in full industrial electrification.

“We support a Spanish company called Roca, which makes ceramics, and we have electrified all their process. It’s an interesting case, and we are starting to see many more cases like that. It’s not only because of a sustainability driver. It’s also not to depend so much on gas,” she said.

Energy-intensive European companies have grown acutely sensitive to gas price volatility since the disruption of Russian supplies, and EU subsidies now support electrification investments. Espada said energy security and risk diversification are now the primary drivers of boardroom energy conversations, making it easier to win the case for efficiency investments at the CFO level.

Artificial intelligence (AI) data centers can flex their power consumption in real time without disrupting computing workloads, a capability that could transform them from grid burdens into stabilizing assets for electricity networks.

A consortium of technology and energy companies has demonstrated for the first time in the UK that an AI data center can respond to grid signals within 30 seconds, cutting its power draw by more than a third and sustaining reduced consumption for hours when renewable generation is low.

“When we get the signal — middle of the night, lightning strikes — we’re able to reduce power within 30 seconds by about 35%,” said Varun Sivaram, chief executive of Emerald AI. “Later this year, we will have taken this technology from lab bench invention to commercialization at a commercial data center in just two years.”

“If you can throw more electrons at a fixed-cost system and you don’t need to build more infrastructure, then the rates come down for everyone else,” said Steve Smith, president of National Grid Partners. “You bring them on, they’re paying towards your fixed costs, and therefore it’s good for rate payers.”

The trial, jointly conducted by Emerald AI, National Grid, Nvidia and Nebius, ran for five days in December 2025 at Nebius’s new data center in London, according to a press release issued on March 2. It tested Emerald Conductor, Emerald AI’s software platform, on a cluster of 96 Nvidia Blackwell Ultra GPUs. National Grid sent more than 200 simulated grid stress events to the site without advance warning, including sudden generation losses, demand spikes and extended periods of low wind.

The system passed every test. It cut power by up to 40% during peak-smoothing events, including half-time surges during major football matches. It shed more than 30% of its load in roughly 30 seconds during a simulated system stress event, and followed load-reduction requests for up to 10 hours, all while protecting critical AI workloads running at full throughput.

The trial’s significance extends beyond a single facility. As the UK prepares for more than 6 gigawatts of data center deployments on the grid by 2030, the companies estimate that rolling out this technology across the sector could add more than 2 GW of flexible capacity back to the grid when needed, easing connection queues and supporting the integration of low-carbon generation.

The announcement is the latest in a series of moves by major AI infrastructure companies to position data centers as grid-friendly assets rather than sources of strain. National Grid and Nvidia are both investors in Emerald AI, which was founded in November 2024.

Later this year, Nvidia and Emerald AI plan to launch the Aurora AI factory in Manassas, Virginia, a 100-megawatt facility set to become the world’s first fully power-flexible AI data center.

Smith said the key to unlocking this flexibility was persuading data center operators that not all workloads are equally urgent. What initially seemed like an insurmountable constraint turned out to be a solvable engineering problem once operators were shown the physics of electric power networks and the costs of inflexibility.

Three levers, one brain

The panel on decarbonizing data and AI’s energy footprint took place at the 12th annual Sustainability Week, organized by Economist Enterprise in London. The discussion was moderated by Vijay Vaitheeswaran, global energy and climate innovation editor of The Economist, and supported by Nebius and National Grid.

Alongside Smith and Sivaram, the panelists were Josh Parker, head of sustainability at Nvidia; Anuja Ratnayake, emerging technologies executive at Electric Power Research Institute (EPRI); and Daria Mukhortova, head of sustainability at Nebius.

Sivaram described three methods Emerald AI uses to achieve power flexibility without compromising service levels:

• Pausing or slowing low-priority workloads, such as a fine-tuning run not needed until the following week.

• Migrating workloads between data centers. In a prior demonstration with EPRI, Emerald moved a workload from Virginia to Chicago across two Oracle data centers at a latency penalty of just 10 milliseconds, imperceptible to the end user.

• Deploying on-site batteries, generators or fuel cells to absorb or release power independently of compute loads.

“There are autonomous AI agents, the secret sauce, that are talking directly to Nvidia GPUs and to the job scheduler, understanding autonomously that they have to meet the lightning strike while protecting the integrity of the workloads,” Sivaram said. “That’s an optimization problem no human can solve, but AI can.”

Nvidia’s interest in grid flexibility goes beyond its investment in Emerald AI. Flexible data centers can be connected to the grid faster, a real competitive advantage in a supply-constrained energy market. Google’s experience in Michigan and other Midwestern states showed that offering demand flexibility can secure faster grid connection approvals. Nvidia has also developed templates to help its customers adopt flexibility practices.

“Load growth historically hasn’t been a bad thing in terms of economic development, but also in terms of sustainability, especially when we’re on the threshold of clean energy being more economical than traditional fossil fuels,” Parker said.

Flexible AI demand can absorb surplus renewable generation when it is available, raising the value of variable energy sources and making data centers a critical component of virtual power plant ecosystems, where AI helps match variable supply with variable demand.

AI’s peak load problem

Ratnayake said EPRI’s engagement on AI data center flexibility stems from a near-universal pain point among its 500-plus members across 45 countries. The traditional timeline for connecting a large new industrial load to the grid is seven to ten years, but data centers need access within two to three years.

“Almost every one of our members is seeing a pain point right now from the growth of large loads that is making them pause and go: how can we do better?” she said. “We think of it as creating a common language between these two major industries. When somebody says ‘be flexible,’ we know what flexibility means.”

On March 23, EPRI launched Flex MOSAIC, a uniform flexibility classification framework developed through its DCFlex initiative in collaboration with more than 65 utilities, system operators, regulators, hyperscalers and technology providers.

Announced at CERAWeek in Houston, the framework categorizes grid needs into standardized classes, allowing data center designers to select configurations that meet specific local grid requirements and accelerate connection approvals.

A second workstream focuses on interconnection requirements specific to AI factories, a category of facilities whose energy consumption profiles differ sharply from those of conventional data centers built before 2023.

“The new AI factories have a very different profile shape, which is driving additional requirement sets that we need to get into the interconnection requirement documents,” Ratnayake said.

Mukhortova offered a concrete example of how AI infrastructure can contribute to local energy systems rather than simply draw from them. Nebius owns a data center in Finland where a heat recovery module, built into the cooling infrastructure, captures server waste heat and feeds it into the municipal district heating system.

In 2025, the arrangement reduced local heating prices by 10% and avoided 4,000 tonnes of CO2-equivalent emissions, equivalent to removing more than 900 gasoline-powered vehicles from the road.

“AI infrastructure does not necessarily have to be just a consumer. It also needs to be mindful about the grid constraints,” Mukhortova said.

The question of whether AI’s energy demand will overwhelm grids divided the panel.

Ratnayake described what she called the “chunky peanut butter versus smooth peanut butter effect”: overall electricity demand growth has been relatively flat for a century, but AI data centers are concentrating that growth into specific industries over a compressed 15-year window, creating the appearance of a crisis.

Smith said AI’s downstream efficiency gains in logistics, air conditioning, materials science, and energy production could make it a net negative for total demand over time, once model training is complete.

“We don’t have an energy problem per se. We have a peak load problem,” Sivaram said. “These AI data centers can actually be great citizens that lower your bills, make the world more sustainable, and make it possible for us to have innovation in a very community-friendly way. They should be the heroes, not the villains of your community.”

The path to that outcome, panelists agreed, runs through tighter coordination between AI companies, utilities, chipmakers and regulators, and a willingness to treat grid flexibility not as a constraint but as a commercial opportunity.

Companies deploying AI agents in their workflows face a fundamental question that has nothing to do with technology: can the people who rely on those agents actually trust them?

One UK travel firm has found an answer, and it starts with treating an AI analyst the same way you would treat a new hire fresh out of university.

“If you were hiring a human analyst fresh out of university, you would spend the first few months guiding them through your business data: these are the columns, the KPIs and the measures to use in this context,” said Aman Bhattarai, EMEA Customer Success Lead at ThoughtSpot.

“We need to apply exactly the same approach to an AI analyst. Once it understands your business language, it is like getting eight to 10 years of experience in a couple of weeks,” he said.

He added that the principle applies universally, regardless of which AI analytics tool an organization uses. He was speaking about easyJet holidays, a UK packaged holiday provider and a client of ThoughtSpot, where the approach has helped transform a five-person data team into an operation serving 350 employees daily, while the company doubled its revenue without significantly growing its headcount.

EasyJet holidays is a fully owned subsidiary of easyJet, the budget airline, and operates as a packaged holiday provider of around 400 employees. It generated revenue of £1.3 to £1.5 billion (about $1.6 to $1.9 billion) in its latest fiscal year and made around £250 million in profit.

When ThoughtSpot began working with easyJet holidays in early 2023, the holiday firm's data team numbered just five people and was being overwhelmed by a surge in post-COVID travel demand. It could not keep pace with dashboard and report requests coming from around 200 employees at the time.

ThoughtSpot was introduced first with a single use case in the trading team, helping analysts optimize pricing and booking routes. Within six months, adoption had grown from 10 users to 100. It now covers 350 of the company’s roughly 400 employees. In January 2025, easyJet holidays retired every other analytics tool and became a fully ThoughtSpot-run business.

The commercial impact has been significant. Without substantially increasing the data team, which grew from five to around 10 people, the company doubled revenue from roughly £500 million to £1 billion. Of the 350 active users, 100 query the platform’s AI analyst, Spotter, every single day.

They ask questions such as which hotels were most profitable last week or what the average selling price is across different routes, questions that previously required a request to the data team.

The CEO of the easyJet group no longer waits for a daily report from the data team. He opens ThoughtSpot on his mobile phone each morning, with dashboards refreshed every 30 minutes, to check metrics including flight delay rates across the network.

Negotiating live

The session, titled “Analytics everywhere: How easyJet holidays put AI-driven insight in every team’s hands,” was presented at the AI & Big Data Expo, part of the TechEx conference, in London. It was delivered by Jane Smith, Field Chief Data & AI Officer at ThoughtSpot, and Bhattarai.

One of the most vivid illustrations of the transformation came from the trading floor. Before the overhaul, trading managers traveling to destinations such as Greece to negotiate contracts with hotel chains, including Hilton and Marriott, would prepare PowerPoint decks, spreadsheets, or PDF dashboards summarizing 12 months of booking data.

That process has been replaced entirely by live negotiation on mobile devices. Managers now pull up live boards, refreshed that morning, showing real-time booking volumes, lead data and performance metrics for each hotel partner.

“Every single member of the trading team, supply chain or marketing team has not asked for a single dashboard from the data team in the last three years,” Bhattarai said. “The end users, who have been enabled and trained, simply go and ask questions and build their own reports and dashboards.”

The reach of analytics extended to the customer service desk as well. When a customer calls in, a service agent can instantly query that person’s full booking history, any flight delays they have experienced and past cancellations, all in real time.

Bhattarai said achieving that level of adoption required deliberate investment in data literacy. Business users in each domain, whether trading, supply chain or marketing, were trained on which data columns and measures were relevant to their specific function rather than being given generic analytics instruction.

Smith said the AI analyst draws not only on structured enterprise data but also on knowledge stored in collaboration tools such as Slack, Microsoft Teams and Confluence, giving it a broader base for answering questions.

The data team, freed from building reports, pivoted to higher-value work: sourcing data from customer relationship management (CRM) systems, net promoter score (NPS) tools, customer service engagements and booking platforms, and building the semantic models that underpin both executive dashboards and AI agents. The data team has not built a single ThoughtSpot dashboard itself.

Smith said the shift at easyJet holidays illustrates how far an organization can move along the analytics maturity curve.

“It is the future of data and AI teams. We can almost see it start to happen now at easyJet,” she said.

The change was not painless. Bhattarai said the hardest part of the transformation was not the technology but the cultural shift. Data analysts who previously owned dashboards as personal products felt defensive when those were challenged or replaced. Business users had to be persuaded they no longer needed to rely on a colleague in the data team to get the right answer. They could trust themselves, or an AI agent, to do it instead.

He said that managing this change across large enterprises had been the biggest struggle for him and his team.

Coaching the agent

Trust is now the defining challenge of the agentic era, more so than speed or scale, which the travel firm had already solved. Bhattarai said once an AI analyst is properly onboarded with business context, the returns are asymmetric: what takes a human analyst months to absorb can be instilled in an AI agent within weeks.

Smith mapped the broader industry trajectory: from operational reporting to visual dashboards, then search-based analytics, then Gen AI and agentic tools, and finally fully autonomous analytics where agents not only surface insights but take action.

She cited Gartner's forecast that half of all business decisions will soon be automated by agents, and McKinsey's projection that a huge amount of focus, budget, and energy is shifting toward agentic AI.

For organizations still deciding where to begin, she recommended abandoning the traditional two-year, top-down exercise of data cleansing, governance frameworks and product roadmaps. Instead, identify one or two use cases at the edge of the business where data directly touches users and start there.

She said if building out that product reveals gaps in governance or weaknesses in the underlying data infrastructure, those can be fixed along the way. The approach generates visible proof points, secures stakeholder support and avoids the risk of a lengthy project that delivers nothing tangible in the short term.

Nearly all enterprise artificial intelligence (AI) pilots are failing. Not because the technology does not work, but because organizations are deploying it in the wrong order, with the wrong people in charge and without the training structures needed to make it stick.

The problem is not agentic AI itself. It is the deeply human question of how companies introduce it, and a growing body of evidence from more than a thousand real-world deployments suggests that most are getting the fundamentals badly wrong.

“95% of pilots, where people were trying agentic workflows and building AI agents, are not working. That’s a crazy high number, considering the amount of money and effort going into this technology,” said Faateh Dhillon, AI specialist at Dust, an agentic AI platform. “There is more than a 97% downfall on those hand raises of colleagues using AI agents, and I really want to talk about why this gap exists.”

Dhillon asked the audience two questions by show of hands: how many had personally tried to build an AI agent, and how many could say 90% of their colleagues were doing the same. The first drew many raised hands; the second drew almost none, a gap he described as the central challenge facing enterprise AI today.

He observed a clear shift from late 2025 into early 2026, as industry conversations moved from “what is agentic AI?” to companies actively building agents at scale. The pace of that transition is accelerating, and organizations that have not yet established a coherent adoption framework risk falling significantly behind. Without one, investment in the technology tends to stall at the level of individual experimentation and fails to translate into measurable business outcomes.

The Paris-based Dust has worked with more than 1,000 companies over three years, including e-commerce software firm Clay, code editor Cursor and security compliance platform Vanta, deploying AI agents across all departments, primarily with non-technical employees who do not know how to code or build automation workflows.

That experience has given the company a clear view of where most enterprise rollouts break down and what separates the organizations that reach near-total adoption from those that stall.

Give power to people

Dhillon delivered the session “Getting to 90%: The AI Adoption Playbook for Enterprise” at the AI & Big Data Expo, part of the TechEx conference, in London. The talk drew on data from Dust’s customer deployments to offer a step-by-step framework for achieving organization-wide adoption of AI agents.

The core argument is that most companies approach AI rollout in the wrong order.

“The secret to success lies somewhere in the middle, but in a very particular order. It’s bottom-up first, eventually with the top-down structure layer that follows,” Dhillon said.

Top-down adoption, where C-suite executives mandate AI usage, is a common starting point, but it tends to produce compliance rather than genuine engagement.

“Executives do need to lead the charge, but in a certain way. You need to showcase to people that they have the capacity to experiment and to fail,” he said. “Don’t put it in a box, don’t only give it to three people and then say that you’re doing the right job.”

Bottom-up adoption, giving every employee, regardless of seniority, permission to experiment, is the right first move. With this approach in place, companies typically reach 50–60% internal adoption. But without layering in top-down governance around the 70% mark, further scaling becomes very difficult.

Dust itself fell into this trap: five employees independently built the same agent, leaving colleagues unsure which version to use for routine queries.

On the question of who should build AI agents, IT departments are frequently the wrong answer. Most organizations default to IT controlling who builds on AI platforms, distributing licenses, and owning deployment, a model that concentrates power in the wrong hands and actively harms adoption.

“This kind of technology needs to be put in the hands of every single individual, not a black box in one room. No one should have to come to IT saying, ‘Can you build me this AI agent?’ That salesperson needs to be enabled to do this themselves,” Dhillon said. “IT’s role has to become that of the facilitator, not the one that takes the power.”

He cited a fraud detection team at one of Dust’s largest customers that built an AI agent without any engineering or IT involvement, saving approximately $100,000 in five days.

“The goal is not to show you the monetary value. The goal is who builds this. The builders were not the engineers or the IT team. The builders were the fraud detection team themselves,” he said.

Make training non-optional

A recurring failure point is training, or the lack of it. Organizations that made AI tool training mandatory saw more than 70% of staff complete it, while optional training consistently produced far lower participation and higher dropout. The difference in sustained adoption between the two approaches is stark.

“Optional training sometimes does not have that effect. People experiment, they don’t know what they’re doing, they don’t have anyone to ask, and they simply stop using the technology,” Dhillon said. “If you have to impose one thing on your team, one hour a day of the week, please do the training.”

Mandatory training matters more for AI agents than for previous enterprise tools because agents represent a genuinely new paradigm. Without structured onboarding, employees unfamiliar with prompting and the differences between large language models (LLMs) quickly become overwhelmed and disengage before the technology has a chance to prove its value.

On choosing who leads adoption within teams, Dhillon cautioned against the instinct to pre-select champions.

“The people that really ended up creating interesting use cases were not the ones we were suspecting. Let the champions emerge organically, because that will show who is actually understanding and willing to spend time with this technology, versus someone you picked who might not be the best person,” he said.

AI adoption is fundamentally a cultural challenge, not a technical one. “It’s a cultural shift, a mindset shift, because it’s giving you a lot more processing power than we’ve had before,” he said.

On measuring success, Dhillon drew a clear line between early-stage and mature-stage metrics. At low adoption levels, organizations should track activity: how many agents have been built and what share of staff are using them.

Once adoption crosses roughly 70%, the focus should shift to outcomes.

“If I had a customer success manager handling $2 million of accounts, can they now handle $3 million because of the agents in place? You start connecting the usage of that agent with actual outcomes inside your organization,” he said.

E-commerce platform operator Mirakl reached 55% AI agent adoption before introducing top-down governance. Within 30 days, adoption rose to 90%, with nearly every employee using agents daily. In one Dust customer survey at 80% adoption, 90% of employees said losing access to the platform would seriously harm their work, a sign that the technology had become truly indispensable.

Proving that artificial intelligence (AI) delivers tangible business value, rather than merely cutting costs, has become the defining challenge for executives scaling autonomous systems in 2026.

Organizations that frame AI returns in terms of output metrics, platform longevity and data infrastructure gains are more likely to secure sustained investment than those that rely on headcount savings alone.

“Return on investment (ROI) is not a cost-succession feature,” said Riccardo Calliano, Vice President of Finance for GenAI Commercial Investments at GSK Pharmaceuticals. “ROI is a business metric. If I have an AI program that helps me better target my customers, my ultimate impact is on sales. How much have I increased my sales versus my target because of my tool?”

Calliano outlined three ways to evaluate ROI:

• Tying it directly to a business outcome such as revenue or customer conversion.

• Treating AI as a platform asset, similar to how enterprise resource planning (ERP) systems are assessed in corporate finance, where upfront investment enables future use cases at marginal cost.

• Recognizing the data engineering work underpinning an AI rollout as an asset that reduces the cost of subsequent applications.

“If you developed a new generative AI (GenAI) assistant with deterministic and non-deterministic workflows, you prove it works, and then plug in additional use cases at just a marginal cost,” he said.

Divyansh Saxena, Vice President of AI & Data at HomeServe EMEA, noted that the cost of building a proof of concept (POC) has fallen sharply. Two years ago, the same process could take an entire quarter. Today, it can be done in 1 to 2 weeks.

“If you want to prove something quickly to your business, it shouldn’t take more than a week or two to build a quick POC based on the current frameworks,” Saxena said. “ROI comes later once you have proven things, but the amount of investment required to prove or disprove things is much less in 2025 to 2026.”

On productivity, Calliano cautioned against inflating the numbers. “When we’re talking about productivity, we should talk about extracted value, not saving 10% of my time and adding it all up,” he said. “If you’re extracting time, are you really eliminating activities? Are you reallocating people to other activities so that the process is more efficient?”

Franny Hsiao, EMEA Leader for AI Architects at Salesforce, warned that organizations failing to anchor AI programs to a clear north star risk accumulating the wrong kind of complexity. “We’re creating even more agentic silos and technical debt,” she said.

GSK is one of the world’s largest pharmaceutical companies, operating a global AI program that combines machine learning and GenAI into composite solutions designed to optimize commercial investment. HomeServe EMEA is the European arm of the property repair and insurance membership group, currently rolling out AI agents for internal customer operations.

Salesforce provides cloud-based customer relationship management (CRM) and AI platform software to enterprises globally.

Messy data, failed pilots

The panel, “From Prototype to Production: The Journey of AI in Autonomous Systems,” was moderated by Ravi Jay, Vice President and Global Head of Agile Center of Excellence (CoE) at Sanofi, at the AI & Big Data Expo, part of TechEx, in London.

Panelists included Calliano, Saxena and Hsiao alongside Christine Foster, General Manager for AI and Automation at Experian UK & Ireland.

On moving AI from pilot to production, Calliano described three tests a pilot must pass: desirability, feasibility and value. Piloting works well under an agile methodology with continuous minimum viable products (MVPs), but the real challenge begins at scale.

“The real value gets delivered when you move to scale up, and scaling a tool is very challenging,” he said. “The value to the customer gets impacted by how many customers you have, how they access your tool, your infrastructure.”

Large organizations operating across multiple markets must also balance agile and waterfall development methodologies simultaneously.

“The biggest barriers are really related to organizational barriers,” Calliano said. “It’s not really technological barriers.”

Hsiao identified a structural flaw that undermines many pilots. Most customers, she said, build their pilots on simplified data and workflows, without accounting for the messy, complicated reality of enterprise data.

Governance and observability must be built in from the start, not retrofitted after deployment. Fear of job displacement is also one of the most common reasons pilots stall.

“A lot of pilots fail when people don’t adopt your technology because they are afraid that their jobs are being taken away,” Hsiao said. “Providing that psychological safety first is really important.”

Foster pushed back on the casual use of “human in the loop” as a reassurance mechanism.

“People simplify human in the loop probably too much,” she said. “Sometimes you might have somebody explain a use case and say, ‘but don’t worry, it’s human in the loop.’ It’s quite frankly a lazy way of telling you it’s going to be okay.”

She proposed a three-part framework for evaluating human involvement: at design time, at runtime inference and at the observability layer.

“At runtime, are you expecting a person to keep up with the speed of 100,000x processing?” Foster said. “Think about that one.”

Responsibility by design

The appropriate level of human oversight also depends on industry context. HomeServe’s lending operations are regulated by the Financial Conduct Authority (FCA), which places strict limits on automated financial advice. Consumer apps such as fitness or gaming platforms face no such constraints.

Saxena pointed to LangGraph, an open-source agentic framework built to incorporate human-in-the-loop functionality, as a tool available for regulated deployments.

On responsible AI, Foster drew on a Canadian engineering tradition in which graduates receive a ring made from metal from a collapsed bridge, a physical symbol of professional accountability.

“That responsibility is not just some layer in the tech stack; it’s not something that you get to buy from a vendor,” she said. “It has to be designed right throughout.”

Hsiao agreed, stressing that the standard for responsible behavior is not fixed.

“It’s not something you check off on a checklist, and it’s also not something you do at once and done,” she said. “What responsible means for different people, for different groups, different societies — that concept is ever evolving. It has to be part of your DNA, basically ingrained into the culture.”

Foster added that the UK’s principles-based regulatory approach offers a competitive advantage. “The regulators ask for proof that you’re adhering to the principles, not promises,” she said.

During audience questions, Saxena explained how large language model (LLM)-as-a-judge technology can address compliance at scale. A compliance officer required to review lending-related phone calls cannot realistically listen to more than a fraction of the volume. An LLM deployed as a judge can process all 1,000 calls and surface potential violations.

“Where LLM-as-a-judge comes in is the scalability aspect,” Saxena said. “You can have an LLM go through 1,000 of them and give its opinion.”

Calliano added that the human reviewer’s role then shifts to high-risk edge cases.

“The person will not be drowned in tons of data but will focus on specific high-risk activities already highlighted,” he said.

Looking ahead, Hsiao called for a shift toward “frugal AI,” advocating for batched workloads and smaller specialized models over continuous real-time inferencing at scale. Calliano envisioned composite AI architectures that combine machine learning, GenAI, and agents tuned to specific business problems, especially in accuracy-critical sectors such as pharmaceutical R&D.

Saxena cautioned that as AI systems grow more capable, understanding the “why” behind model outputs is becoming harder, making experienced AI practitioners in the boardroom an increasingly valuable guide between technical possibility and business need.

Schneider Electric will present its latest artificial intelligence (AI) infrastructure solutions at Datacloud Global Congress next week, including validated AI factory blueprints for NVIDIA’s GB300 NVL72 platform and the Grace Blackwell Ultra architecture.

The energy technology company, which employs 160,000 people across more than 100 countries, will demonstrate liquid cooling systems, 800-volt direct current (VDC) power architectures, digital twin capabilities and end-to-end software tools designed to help operators build and run large-scale AI data centers.

The showcase comes as AI infrastructure investment accelerates sharply. Nearly $3 trillion of AI-related spending is expected to flow through the global economy by 2028, according to Morgan Stanley Research, while Gartner forecasts worldwide AI spending will exceed $2.5 trillion in 2026 alone.

Several Schneider Electric executives will lead keynote panels and technical sessions at the congress, which runs from June 1 to 4 in Cannes, France, and will also co-host an invitation-only executive briefing with NVIDIA on the future of AI-driven infrastructure.

“AI is fundamentally reshaping the future of digital infrastructure, creating new demands around power, cooling and resiliency at unprecedented scale,” said Marc Garner, Global President, Cloud and Service Provider Segment at Schneider Electric. “At Datacloud Global Congress, we will demonstrate how collaboration across the ecosystem is enabling the next generation of AI factories and helping organizations build scalable, resilient and sustainable infrastructure built for the AI era.”

Neoclouds, cooling, 10MW

On June 2, Frédéric Godemel, Executive Vice President of Energy Management Business at Schneider Electric, will join executives from Oracle, DATA4, QTS Data Centers and CBRE for a keynote panel on how the data center ecosystem is keeping up with AI demand, covering the rise of neoclouds, European competitiveness and the gap between hyperscale and enterprise requirements.

Also on June 2, Thierry Chamayou, Vice President of Cloud and Service Providers in EMEA at Schneider Electric, will join a panel with GreenScale, Trench Group, Kao Data, JSM Group and Solar Turbines to explore how operators can de-risk energy investments through smarter project structures and closer collaboration with utilities and government.

On June 1, Sébastien Cruz-Mermy, Vice President of Datacenter Innovation at Schneider Electric, will lead a technical session on AI factory infrastructure, focusing on ultra-high-density rack design, next-generation DC power delivery and resilient cooling. Schneider Electric and NVIDIA will also co-host an invitation-only executive briefing on NVIDIA’s 5-Layer Cake framework and the DSX Blueprint, supported by digital twin tools.

At Stand 122, highlights will include the MCDU-70, a 2.5-megawatt coolant distribution unit (CDU) from Motivair by Schneider Electric that can scale to 10 megawatts and beyond, as well as EcoStruxure IT data center infrastructure management (DCIM) software, NVIDIA Omniverse integrations, and NVIDIA Reference Designs for the GB300 NVL72 platform.

The company’s goal is to help customers design, simulate, build, operate and maintain AI-ready infrastructure at scale, with energy efficiency and sustainability as central requirements.

Agentic artificial intelligence (AI) systems may perform flawlessly in controlled demonstrations, but deploying them across large organizations exposes a set of challenges that most vendors and developers have yet to fully address.

From ensuring systems can be shut down instantly to managing workflows that run for months, the gap between a proof of concept and enterprise-grade deployment is wider than it appears.

“They all work well in demos, but when you talk about enterprise adoption, it’s a totally different view and lens that you need to put on,” said Amal Makwana, Senior Vice President and Engineering Manager at Citi.

“If it falls at 2 a.m., something’s wrong. The first set of questions to be asked is: Can I shut it down? Does it have a kill switch? Can I throttle it? Those are the things that enterprises look for,” he said.

He said the eventual architecture of enterprise AI will likely see every business department running its own dedicated agent.

“Every department has an agent, and then there is a super agent who’s coordinating between everybody,” Makwana said.

At an event, he outlined the full range of production requirements that enterprises must demand before deploying agentic AI at scale. The list includes observability tools for root cause analysis, robust integration with legacy applications, customization and extensibility, and state management for regulatory compliance.

Developer experience also matters. The quality of documentation, the availability of debugging tools and the learning curve of any given framework all factor into enterprise readiness.

No single agent can be an expert across all business domains, which makes a coordinating orchestration layer essential. In financial services, long-running workflows can span multiple months, making state management especially critical.

“In the financial world, we love building wrappers around everything,” he added.

Citi is one of the world’s largest financial institutions, operating across more than 160 countries and serving corporate, institutional and retail clients. Makwana leads engineering delivery teams focused on the bank’s technology transformation programs.

AI Ops goes autonomous

Makwana's presentation, titled "Deep Dive: Agentic Autonomous Systems," was delivered at the AI & Big Data Expo, part of the TechEx series of technology events held in London. The session drew a mixed audience of engineers, enterprise architects and technology managers.

He described two categories of emerging use cases: AI operations (AI Ops), which automates and optimizes IT service management, and four autonomous professional roles he expects to become mainstream in the near term.

On AI Ops, Makwana described capabilities including historical data analysis, anomaly detection, performance monitoring, cross-application data correlation, optimization suggestions and automated task execution.

“Next time there’s an increase in transactions because there’s a sale on Christmas, the system will automatically increase your memory and compute power to make sure that everything is okay,” he said.

AI Ops is distinct from two related disciplines. DevOps focuses on faster software delivery through automated continuous integration and continuous delivery (CI/CD) pipelines, while machine learning operations (MLOps) govern how ML models are selected, trained, tested and deployed into production applications.

Beyond IT operations, he outlined four roles in which autonomous agents are set to handle tasks currently performed by humans:

• A travel agent that plans and books trips based on known preferences.

• A financial advisor that monitors portfolios and rebalances them automatically.

• A developer agent that writes code to organizational standards with minimal prompting.

• An expert route planner that factors in constraints and transport modes.

He said choosing the right framework requires careful analysis. The key criteria include the complexity of the use case, the desired level of human oversight, latency and performance requirements, and deployment infrastructure.

Security and regulatory compliance are particularly important in financial services, where firms must adhere to rules such as the General Data Protection Regulation (GDPR). Community support and cost round out the checklist.

Agents will rarely operate in a greenfield environment, meaning integration with existing legacy applications is a non-negotiable consideration. For organizations in the early stages of experimentation, Makwana suggested open-source frameworks as a practical starting point before committing to commercial platforms.

Decade, not a year

Understanding what qualifies as a truly agentic AI system, as opposed to a simple AI agent, is a distinction Makwana said is frequently misunderstood.

“Agents are not agentic AI. They are the building blocks of agentic AI,” he said.

He described four observable traits that define a genuinely agentic system: adaptability, agency, autonomy and persistence. He said that any system exhibiting those four traits can be classified as agentic AI. The determination requires no technical knowledge; observing how the system behaves is enough.

He traced the evolution of such systems through four stages, using a medical analogy. The first is a basic reactive large language model (LLM), capable of answering general questions. The second adds retrieval-augmented generation (RAG), giving the model access to verified, trusted data sources.

The third stage embeds predefined workflows, enabling the system to coordinate tasks rather than merely advise. The fourth and final stage is a fully agentic system capable of dynamic planning, goal persistence and coordinated action across multiple agents.

Five core components are required to reach that final stage:

• Intent and awareness: allowing the system to pursue goals rather than wait for specific instructions.

• Intelligence and decision-making: to reason through options within policy constraints.

• Execution and learning: enabling step-by-step action refined by feedback.

• Coordination and control: to orchestrate multi-step plans across tools.

• Governance and reliability: ensuring the system operates within defined boundaries and can be explained.

“It has to operate within a set of guardrails. It needs to know its boundaries, and it has to work reliably. We need to be able to explain what it does and how it does it,” Makwana said.

On the broader state of the industry, a late-2025 McKinsey report showed that nearly two-thirds of organizations remain in the experimenting or piloting phase. While 64% said AI helps their operations, only 39% reported a measurable impact on earnings.

Half identified workflow redesign as a critical success factor, and 80% said AI should drive growth and innovation, not only cut costs.

“More EBITDA [earnings before interest, taxes, depreciation and amortization] impact will be seen if you look at AI in a more holistic way, and not just as a tool,” Makwana said. “You need to redefine the workflows.”

Three interoperability protocols are gaining traction in the agentic AI space. Model Context Protocol (MCP) standardizes how LLMs access external tools and APIs. The Agent-to-Agent (A2A) protocol, initiated by Google, allows agents to advertise their capabilities and communicate directly. The Agent Communication Protocol (ACP), proposed by IBM, achieves similar coordination via HTTP rather than direct real-time connections.

He stressed that the winner in the coming decade of AI adoption will not be whoever moves fastest.

“This will not be the year, but the decade of AI and agents, and the winner would be the one who doesn’t adopt fastest, but adopts wisely,” he said.

Fear of job displacement and confusion about the scale of change are the two biggest reasons workers resist artificial intelligence (AI), and companies that fail to address these concerns before rolling out new technology are setting themselves up for failure.

Acknowledging the fear is not enough. The real challenge is listening to what that fear is actually saying, and then breaking the change into pieces small enough for people to swallow.

“The question we need to be asking is why, and we have to listen to their why,” Jena Miller, a specialist in global transformation and adoption, told TechJournal.uk in an interview in London. “Are they afraid they are going to be replaced, or do they think the change is too big?”

“When we break it down and simplify it, and say this portion of AI is what we’re trying to introduce, you’re not swallowing the elephant, you’re eating a piece of mandarin. It becomes a lot easier to swallow,” she said.

Miller said the answer lies not just in communication but in ownership.

“When people as a collective believe in what you are trying to do and they own it themselves, your audience and your message become bigger,” she said. “There is a lot more power in a voice of 10 than in the voice of one.”

The pattern holds across sectors. She has spent her career in manufacturing, consumer goods and automotive industries. Companies are always trying to optimize and compete, but how they get there matters as much as the destination.

She said the most common mistake is trying to drag a team along rather than creating conditions that make the team pull leadership forward with their own ideas.

Clarity, alignment, inclusion

Miller was speaking at the AI & Big Data Expo, part of the TechEx event series, held in London. Her presentation, titled “Designing Transformation People Will Actually Adopt,” drew on her experience delivering large-scale change programs across global organizations.

The central argument of her talk was that technology is rarely the problem. Research from McKinsey, Prosci and the Harvard Business Review all points to the same finding: adoption, not technology, is the number one indicator of transformational success. McKinsey found that 70% of all change programs fail to meet their goals, not because the technology does not work, but because people cannot or will not use it.

“When people don’t trust or understand the change you are trying to introduce, technology can’t even begin to scale to its potential,” Miller said.

To address this, she introduced what she calls the adoption equation: three conditions that must exist simultaneously for change to stick. They are clarity, alignment and inclusion.

(1) Clarity is not about key performance indicators (KPIs) or deadlines.

“It’s the meaning. Meaning answers a very personal question: how does this affect my life, how does this affect my work, and how does this connect me to something larger than myself?” She used the example of an airport delay: being told a flight is 90 minutes late is just a metric, and passengers adapt by waiting.

But when a gate agent explains the delay is caused by a cabin pressure issue found during safety checks, the same wait feels entirely different. Passengers understand the why and may even feel relieved.

“People don’t resist change,” she said. “They resist meaningless change.”

(2) Alignment ensures that the proposed solution actually fits the environment in which it has to operate. She described rolling out a standardized workstation: the desk, chairs and core software are the standards. But engineers need adjustable monitor arms, finance needs secure document storage and human resources (HR) needs ergonomic flexibility.

Alignment invites the people closest to the work to help shape how change shows up in their day-to-day environment. It is not a redesign but a local integration that replaces friction with cooperation.

(3) Inclusion goes further.

“People will adapt under pressure,” she said. “People will adopt when they’re included, and ownership replaces resistance.”

She used the car purchase analogy: the engine, frame and safety features are non-negotiable, but inclusion lives in the options.

The long-distance commuter picks comfortable seating and a better sound system; the parent picks easy-to-clean materials and extra storage. Neither buyer is redesigning the vehicle. They are choosing how it shows up in their lives, and that choice creates ownership.

“Clarity gives people the why, alignment gives people the how, and inclusion gives people the ownership,” she said. “When you have all three, change doesn’t just show up, it sticks.”

Hammer on the floor

The clearest illustration of the adoption equation in action came from Miller’s own career. While leading a global rollout of a new manufacturing component she described as standardized, efficient and scalable, she visited a shop floor expecting to see her design working smoothly.

“I saw something that no report would have ever captured. The people responsible for loading that perfect part into the machinery were physically beating it in with a hammer, and that hammer was not part of the process,” she said.

“These weren’t people trying to make the process difficult. They were trained technicians with years of experience,” she said. They were the true experts, desperately trying to make a solution work within constraints that the designers had never considered. Those designers had been working from computer simulations and had never visited the fixture in person.

Rather than defending the standard, Miller brought the technicians together and asked one question: if you could change one thing, what would it be? The group combined their tools, lived experience and expertise to arrive at a solution that was better than the original design. That solution was subsequently rolled out across three other sites that had been quietly struggling with the same problem.

An audience member asked how to manage situations in which employees know a disruption is coming but the details have yet to be announced. She said starting the conversation early is critical. Even when it is too late to alter the solution, giving people a voice in how it will affect them creates a sense of inclusion and makes the transition easier to accept.

On the question of cultural differences, she drew on her global experience to note that not all workforces respond to change in the same way. Her colleagues in Asia tend to want more detail and more of the underlying logic. Americans want the backstory and the why. German colleagues want facts, clear cause-and-effect and a concrete sense of the reward.

Taking time to understand not just the individuals involved but the culture in which they operate is, she said, itself part of the adoption equation.

As AI transforms industries at an accelerating pace, Miller’s framework points to a persistent gap between what technology can do and what organizations actually manage to change, and suggests the solution lies less in better tools than in better listening.

Enterprise artificial intelligence (AI) is generating a new division of labor in the back office. Accountants no longer check contractor invoices line by line. Business analysts no longer manually pull data from disparate systems to build dashboards. AI agents are doing it instead.

The shift is being driven not by better models, but by better data architecture. When an agent knows exactly where to find the data it needs, results move from impressive demos to measurable business outcomes.

"All the accountants who did that in the past manually can just focus on what the cost and benefit of the project is, instead of working on checking the various invoices," said Andreas Krause, Vice President of Customer Advisory at SAP EMEA Data and AI. "The agent assigns project costs exactly to the different metrics, and measures what was agreed from a budget perspective when the project was agreed with the contractor."

Krause gave two examples of AI agents already operating in enterprise environments.

• The first is a customized agent built for a utility company that automates contractor invoice validation for grid maintenance projects. The agent checks whether each invoice meets the required detail levels, compares actual costs against agreed budgets, flags overruns and incomplete entries, and automatically sends an email to the contractor requesting missing information.

• The second is an embedded analytical agent that generates complete data visualizations from a natural language prompt. In an HR use case, the agent analyzed staff terminations by pulling salary grades, headcount and demographic data. It displayed its reasoning step by step, showing which data sources it was connecting and which key performance indicators (KPIs) it would use, before producing a finished dashboard.

SAP provides an agent builder that allows customers to define agent tasks and connect whichever large language model (LLM) they prefer. The company has partnerships with all major LLM vendors, leaving the model choice entirely to the customer.

Data gaps, dead pilots

Krause was speaking at the AI and Big Data Expo, part of the TechEx event series, held in London. His session, titled “Winning the AI Race: Turning Enterprise Data into a Strategic Asset,” examined why so many AI initiatives stall before reaching production.

He said a study found that poor data readiness is the leading cause of AI pilot failures, accounting for roughly 30% of cases in which projects never reach production.

He illustrated the problem with a common business query that requires five years of historical financial data, forward-looking forecasts, and addressable market figures from external systems. An agent that cannot locate and connect all three data types will either return incomplete answers or hallucinate.

“A lot of AI prototypes never leave the prototype stage and go into production. One of the main reasons is poor data readiness. Do we really have access to all the data we need, and is the data at the necessary quality so that it gives back real results?” he said. “When the agent doesn’t exactly know where to access the data, the result will simply not be what you expected, and in the worst case it brings back a hallucination.”

SAP’s own AI assistant, Joule, faces the same constraint. Without clear data access, even an enterprise-grade tool will fail.

SAP’s solution is a three-layer architecture: SAP and non-SAP applications at the base, Business Data Cloud in the middle and AI agents at the top.

“A customer needs to have a unique definition within all SAP applications, but also the non-SAP applications. Business Data Cloud harmonizes this together, so that there is one unique definition of a customer, a cost center or material information,” Krause said. “We have an open data partnership with various vendors. We started with Snowflake, in addition to Databricks, but we opened it up to Microsoft Fabric, Google BigQuery, and more to come during this year.”

Business Data Cloud incorporates SAP Datasphere for semantic modeling and KPI calculations, SAP Analytics Cloud for front-end visualization, and SAP Databricks for AI and machine learning workloads involving non-SAP data. Customers using SAP’s legacy Business Warehouse system can also connect it to the platform.

Krause demonstrated two intelligent applications that combine SAP and external data. A working capital application pulls enterprise resource planning (ERP) financial data alongside external credit risk ratings. Finance teams can model scenarios, estimating the cash flow impact if customers pay within 30 or 60 days, and receive best-case, realistic and worst-case projections.

A second application, built by a manufacturing customer, compares preventive versus corrective maintenance costs. It draws on SAP material and cost data, supplier pricing and sensor data from machinery to identify the optimal maintenance cycle and estimate potential savings from each approach.

Security layer built in

Data security emerged as a significant concern in the Q&A, with participants noting that companies are racing ahead with AI while cybersecurity readiness lags, a pattern linked to the incoming EU Cyber Resilience Act. Krause acknowledged that after poor data readiness, security is the second most common reason AI projects fail to reach production.

For embedded AI capabilities, security is inherited from SAP’s existing application-level controls. For customized AI built on Business Data Cloud, a separate framework applies.

“Business Data Cloud also has the security framework behind there, so that you know who is going to access the information and what type of data they are able to see. We have a knowledge graph where you can apply security down to the data level,” Krause said.

The knowledge graph also governs what each LLM is permitted to query, ensuring models cannot access data beyond their defined scope.

An audience member raised the question of whether banks in countries with strict data regulations should build AI exclusively using locally deployed LLMs rather than cloud-based models.

Krause said SAP is progressively building sovereign cloud environments for countries with data residency requirements, designed for public sector entities and regulated industries. These ensure data does not cross national borders. Some countries are also developing their own LLMs for exclusive domestic use, and SAP’s open LLM framework allows these to be connected to its platform.

On implementation efficiency, pre-built data products cover cost centers, materials, customers and other standard business objects. They include validity periods, currency information, and full descriptive metadata, eliminating the manual data modeling that previously required reconciling multiple SAP ERP tables. The reduction in implementation effort can reach up to 81%, growing in proportion to the complexity of the data environment.

SAP plans to expand its library of intelligent applications into industry-specific use cases, with consumer products and retail already in development. Further verticals and additional open data partnerships are expected before the end of 2026.

Knowledge workers are losing 11 hours every week not because they lack tools, but because they have too many of them. Despite four decades of investment in software, apps and automation, the most critical work in organizations still flows through people, and that, according to one of Europe’s leading artificial intelligence (AI) companies, is the real problem.

The bottleneck is not technology. It is agency.

Existing automation handles predictable, repeatable processes well enough, but breaks down the moment a business expands into a new region or changes a workflow. When that happens, humans are left to pick up the slack, checking that automations are working, transferring data between systems and validating outputs rather than doing meaningful work.

“It breaks when a process changes, or it breaks when our business expands into new regions, or we bring in new lines of business,” said Scott Ivell, VP of product marketing at DeepL. “There’s a heavy engineering effort required to keep these systems going, and it’s really too brittle for the world in which we find ourselves now.”

“Our view is this productivity gap, the automation gap, is an agency problem,” he said. “We have to move from this world of using software as a tool to using AI and agents as a coworker.”

Ivell said companies should not wait for their next CIO planning cycle, but instead identify human-intensive processes today and deploy agents immediately. The real competitive advantage, he said, will belong to companies with the most friction-free workflows, where humans stop acting as connectors between systems and become architects instead.

DeepL began as a language translation service and now describes itself as a global leader in applied AI, with over 100,000 enterprise users worldwide.

The Cologne-based company, which counts European heritage and compliance as core to its identity, initially focused on removing friction between languages. It has now turned its attention to a harder problem: the friction between humans and the work they are trying to execute.

Its latest product, DeepL Agent, positions AI not as a tool but as a digital coworker capable of reasoning, planning and acting across an organization’s existing systems, without requiring a single line of code from the people using it.

Nine minutes, four systems

The AI & Big Data Expo, part of the TechEx conference, took place in London. The session, titled "Say hello to your AI co-worker," made the case for agentic automation and included a live product demonstration. Ivell was joined on stage by Ire Adewolu, a senior solutions engineer for agentic AI at DeepL.

Adewolu demonstrated a workflow built for a customer success manager tasked with improving product utilization and revenue across a portfolio of accounts. The conventional approach of consulting a Salesforce expert, reviewing billing data and speaking to a business intelligence team typically takes four to five days. The DeepL Agent completed an equivalent task in under nine minutes.

In the demo, the agent autonomously navigated three to four internal systems using existing single sign-on (SSO) credentials, without any manual development or API configuration. It accessed DeepL’s billing system and found that a customer had purchased 50 licenses but had only three active users, averaging 2.67 per month.

“The agent has seen that a customer has only three active licenses, but has actually purchased 50 licenses,” Adewolu said. “This really reflects that potentially the customer is not gaining value from the product, or maybe hasn’t been onboarded correctly, all of which will pose a significant churn risk.”

The agent then retrieved a relevant onboarding guide from DeepL’s support documentation, drafted a personalized email to the customer and proposed meeting dates by syncing with the user’s calendar, all without human intervention.

The agent is designed to draft rather than send emails by default, preserving a human review step before any customer-facing communication goes out. Users are notified when a draft is ready, and can either edit and send it themselves or authorize the agent to send it on their behalf.

The productivity case becomes clearer at scale. Manually reviewing 100 customer accounts across four or five systems each could take weeks. With sub-agents, which are parallel instances the system can spawn simultaneously, the same volume of work can be processed in minutes. Ivell said a single agent can launch 40 sub-agents to research, analyze and execute in parallel.

“Great humans do one task at a time,” Ivell said. “An agent can spawn 40 more sub-agents to work in parallel, do the research, do the analysis, execute simultaneously. This isn’t just fast, it’s a different dimension in unlocking productivity.”

Workflows can also be scheduled to run daily, hourly or weekly, or triggered via an API, removing the need for repeated manual prompting.

On a short leash

The prospect of an AI agent operating autonomously across enterprise systems raised immediate questions from the audience. Three attendees raised concerns about security, access controls and audit trails, a line of questioning that Adewolu and Ivell said reflects the exact design priorities baked into the product from the start.

“Humans need to move from doers to becoming directors, and success isn’t about giving up control either,” Ivell said. “It’s about having a human in the loop who provides the transparency, the audit trails, and human validation at critical junctures within the workflow.”

DeepL built its agent for compliance with the EU AI Act and ISO standards from the outset, not as a retrofit. Data is encrypted at rest and in transit, and the system operates on a minimum-AI principle: the agent handles the heavy lifting but stops to request human approval for critical decisions or when it is uncertain about the next step.

An audience member raised a concern about agents inheriting the same broad system access as the user operating them, including, in the case of a data scientist, the ability to drop tables in a database. Adewolu said access can be constrained at a granular level. DeepL uses an internal system called Phoenix to manage sensitive customer data, and the agent is given only read-only access to it.

Users can create dedicated service accounts with restricted permissions for each connected application, and the agent is trained to recognize which actions require explicit human sign-off before proceeding.

On audit trails, he said DeepL is developing watermarking for agent actions and already provides reports at the organizational level that show every interaction and every prompt, with timestamps and user attribution. This allows organizations to trace the basis for any decision the agent made, a capability critical for regulated industries where decisions affecting customers, such as insurance underwriting, may need to be explained after the fact.

DeepL Agent also works without pre-built integrations. It opens its own browser and navigates systems as a human would, including those behind a virtual private network (VPN) or firewall, requiring no coding and no lengthy IT rollout.

“True disruption shouldn’t require an 18-month IT project that you need to roll out globally from your CIO,” Ivell said. “The most successful agents operate on your existing stack to navigate the systems that may not have APIs.”

Commands can be issued in any language, making the technology accessible to non-technical staff across multilingual organizations. DeepL said the winners of the next few years will not be the companies with the most AI tools or the most advanced integrations, but those with the most friction-free workflows, where humans have stepped back from acting as connectors between systems and taken on the role of architects instead.

When a customer walks into a hardware store clutching a broken part they cannot name, AI may now be their best hope of finding a replacement. Across retail, banking and financial services, companies are moving well beyond experimentation, deploying artificial intelligence (AI) that solves tangible problems and generating results they can take to the boardroom.

The most successful deployments share a common trait: they were built around a specific business problem, not around the technology itself.

“Most of the time our customers go to our stores not to look for a product but to solve a specific problem. They go and describe their problem to our colleagues: ‘I want to remove all the wallpaper, I don’t know how to do that,’ and then we have to actually come up with the product and how you do it,” said Mohsen Ghasempour, Chief AI Officer at Kingfisher.

“We launched our first public-facing DIY system in December 2023, where you could actually go and ask, ‘I want to tile my bathroom, how do I do that?’” he said.

Kingfisher operates B&Q and Screwfix across more than 2,000 stores in Europe, employing 76,000 people. Beyond the DIY chatbot, it developed Lens, a mobile visual scanner that searches 2.1 million products on diy.com to identify broken or unrecognizable parts. The platform handled over one billion product interactions last year.

Paul O’Sullivan, SVP of Solution Engineering and CTO for Salesforce UK and Ireland, described a similar logic at work in the contact center. After deploying its own AgentForce AI agent on help.salesforce.com, Salesforce achieved 83% first-time resolution across all inbound queries.

“That has relieved a tremendous amount of frontline pressure, enabling us to redeploy people into areas that can drive higher value tasks and ultimately more employee satisfaction,” he said.

The results extend to Salesforce’s customers. Heathrow Airport shifted 70% of phone-based contact center volume to web-based agent interactions. Simply Health, with over 2 million members, reduced claims processing time from 12 minutes to one minute after automating 87% of its claims workflow.

At Citi, Ryan Courtier, SVP and Senior Product Manager for generative AI (Gen AI) platforms, takes a design-thinking approach to finding where AI fits.

“I try to understand from these business leaders: what are the problems you’re trying to solve today? What kind of challenges do you have?” he said. “I try to help them understand this is where AI will enable you to free up very manual, repetitive processes, when you could spend a lot more time on the purpose of what your business is trying to achieve.”

Governance as a launchpad

The panel, titled “Transforming Industries with AI and Big Data: Success Stories from the Frontlines,” took place at the AI and Big Data Expo in London, part of the TechEx Global 2026 conference.

It was moderated by Mark Sage, Executive Director of the Augmented Reality for Enterprise Alliance (AREA) and the Enterprise Data Management Alliance (EDMA), and featured O’Sullivan, Courtier, Ghasempour and Yao Li, Global Chief Product Officer for Data Quality at Experian.

For Courtier, operating inside one of the world’s most heavily regulated banks means governance is not an obstacle. It is a prerequisite. Rather than pitching AI as a technology initiative, he frames every proposal around operational outcomes.

“I treat governance as an enabler. Hallucinations are down to almost zero, and to do that, you need a really close relationship with those teams. Don’t see them as blockers, but as avenues to help you get that value out even faster,” he said.

“If I’m joining a call trying to convince people to use this AI, I feel I’m already setting myself up for failure. I go in and say, ‘I’m going to remove X manual touch points, I’m going to reduce risk by X,’ and that’s often the easiest way to get them on board,” he added.

O’Sullivan argues the governance challenge begins with the large language model (LLM) itself.

“The LLM alone is not enough. When ChatGPT launched in 2022, we all went ‘Wow.’ We very quickly fast-forwarded and saw hallucinations, toxicity and misinformation,” he said. “We consciously focused on trust; trust is our number one value at Salesforce, and we built a trust layer within our core platform to connect to LLMs through a secure gateway, to check for toxicity, bias, and hallucinations, and to provide a predictive score while keeping a human in the loop.”

Ghasempour faced a structural version of the same challenge. Scaling to hundreds of Gen AI agents at a FTSE 100 company would have meant hundreds of separate governance approvals. Kingfisher resolved this by building Athena, a centralized framework that bundles security, legal and compliance checks into a single layer.

“The way we tackled that was to join technology and governance, a central location to apply security measures that gives legal and compliance a bit of assurance,” he said.

Li, whose team at Experian processes 1.2 billion records a month from 700 sources, frames the stakes plainly.

“Where I work really determines ultimately whether AI is truly intelligent or just confidently wrong,” she said. The EU AI Act sharpens the point: violations carry fines of up to 7% of global gross revenue, equivalent to £70 million for a billion-pound business.

Trusted data first

When an audience member asked how AI could support real-time decision-making, Li offered a three-layer answer: trusted data at the foundation, transparency and explainability in the decision process, and scalable governance that embeds institutional knowledge.

“Number one is to start with trusted data at the foundation. If you start with good data, the decision can be faster, near real-time or getting to real time,” she said. “It’s not just about the decision itself, but how do you roll back and explain it to people? You need transparency, explainability, observability and remediation when things go wrong.”

Experian helped a client cut manual data interventions by 50%, resulting in 120,000 data fixes, and helped IKEA reduce duplicate customer records by 15%. It also implemented 150 automated Gen AI workflows to enforce consistent data-quality guardrails at scale.

O'Sullivan described a complementary approach built around confidence thresholds. For fully automated decisions, a confidence score can gate whether an action proceeds. For more complex or sensitive ones, the role of AI is to surface the right information to the right person at the right time.

Salesforce’s Next Best Action feature presents a contact center agent with a relevant offer when a customer with multiple open complaints calls in, rather than automating the response entirely.

The panel cited striking results. Citi is saving 100,000 developer-hours per week by accelerating engineers’ productivity. Kingfisher attributed £18 million in revenue to AI-driven personalization and reported a 15% margin improvement from a pricing optimization project.

“When we talk about a 15% margin improvement, that’s not AI making 15%, that’s people using AI technology to make that impact,” Ghasempour said.

“We’ve seen lots of proof of concepts, pilots, lots of pounds and dollars being chucked at AI that hasn’t really unlocked any business value,” O’Sullivan warned. “I think we’ve got a once-in-a-lifetime opportunity to reimagine how we run our businesses, and the whole business process model will have to be rethought through with agentic AI and workflow automation.”

Closing the skills gap will be central to that ambition. Ghasempour warned that without education, organizations risk both uncritical hype and paralyzing fear of job displacement. O’Sullivan called AI literacy a shared industry obligation, pointing to Salesforce’s £50 million UK investment in hands-on training.

Courtier suggested the most effective reframe comes from Nvidia CEO Jensen Huang: rather than asking which tasks AI can automate, ask what purpose a role is ultimately meant to serve, and let that guide where the technology goes next.

Schneider Electric, a global energy technology company, and its liquid cooling subsidiary Motivair by Schneider Electric have completed a phased delivery of more than $290 million in AI infrastructure solutions at TeraWulf’s Lake Mariner data campus in Barker, outside Buffalo, New York.

The project converts a legacy industrial site into a series of purpose-built AI and high-performance computing (HPC) data centers within a twelve-month timeframe. Upon full buildout, the 750-megawatt (MW) campus is projected to become one of the largest such deployments underway in the United States, supporting artificial intelligence (AI), high-performance computing (HPC) and cloud workloads for anchor tenants Core42 and Fluidstack, the latter backed by Google.

“TeraWulf’s strategy is centered on delivering scalable, energy-efficient infrastructure capable of supporting the increasing intensity of AI and HPC workloads,” said Sean Farrell, Chief Operating Officer at TeraWulf. “By working closely with industry leaders like Schneider Electric and Motivair, we are accelerating the development of AI-ready capacity at our Lake Mariner facilities while reinforcing the strong operational foundations needed to support long-term customer demand.”

“As demand for AI infrastructure accelerates, ‘time to power’ has become a defining constraint on growth. Operators need partners who can bring together advanced infrastructure, services, and expertise in energy technology to underpin large-scale AI data center deployments at pace,” said Manish Kumar, Executive Vice President, Secure Power and Data Centers at Schneider Electric.

“Our partnership with TeraWulf establishes a strategic blueprint for pairing on-site power, AI-enabled automation, advanced liquid cooling, and digital intelligence at a legacy industrial site. We are delivering resilient, efficient, and scalable data center solutions at the speed and scale this AI era demands,” he added.

Strategically located to leverage low-cost, reliable power, the Lake Mariner campus draws from a New York regional power grid whose energy mix is approximately 89% zero-carbon, with substantial surplus power available. TeraWulf is leveraging its partnerships with Schneider Electric and Motivair to drive long-term growth at speed and scale across the campus.

Liquid cooling at scale

The project required Schneider Electric and Motivair to meet demanding construction and operational timelines and to provide technical design and engineering guidance throughout. The build includes Galaxy VX Uninterruptible Power Supply (UPS) units, Galaxy Lithium-ion Battery Systems and NetShelter Racks and Enclosures for power infrastructure.

On the cooling side, Motivair supplied Coolant Distribution Units (CDUs), In-Rack Manifolds and ChilledDoors. Schneider Electric’s EcoStruxure IT Data Center Expert software was integrated to enable advanced monitoring and digital intelligence, while Motivair’s client services team was deployed to anticipate risks, minimize disruption, and maximize the cooling investment.

Liquid cooling has shifted from a niche technology to a core requirement for modern AI infrastructure, as the thermal output of GPU-dense servers far exceeds what traditional air cooling can handle. Motivair’s direct liquid cooling systems enable higher compute density without compromising reliability or energy efficiency.

The Lake Mariner project illustrates a broader trend of repurposing legacy industrial sites, which often carry existing power infrastructure and large land footprints, into next-generation AI campuses. With Core42 and Fluidstack in place and the campus still expanding, TeraWulf and its partners expect capacity and demand to grow in tandem.

Schneider Electric is a global energy technology leader with around 160,000 employees and partners in over 100 countries. The company electrifies, automates and digitalizes industries ranging from buildings and factories to data centers and power grids. Its portfolio includes intelligent devices, software-defined architectures, AI-powered systems, digital services and expert advisory services.

Motivair by Schneider Electric is a leading provider of advanced liquid cooling solutions for high-density computing environments. It serves silicon manufacturers, server OEMs, colocation providers and hyperscale data centers, with a product range spanning chip-level cooling to facility-wide chiller systems.

TeraWulf builds, owns and operates digital infrastructure in the United States focused on AI and HPC. Led by veteran energy infrastructure entrepreneurs, the company delivers scalable and resilient compute capacity for major technology companies across its purpose-built campuses.

The European Central Bank (ECB) is doubling down on a digital euro to counter the creeping dominance of US dollar stablecoins. The UK risks falling behind, leaving consumers no clear right to redeem the foreign-issued stablecoins they hold.

Policymakers are only beginning to grasp that the stablecoin debate is not merely a financial markets question. It is a contest over monetary sovereignty.

“We’re not thinking about it enough. It’s very vital that we realize this isn’t just about financial markets; it’s also about geopolitics, and I don’t think that has really resonated as yet,” said Dr. Lisa Cameron, founder of the UKUS Crypto Alliance and former member of parliament.

“Many of the parliamentarians think, ‘Oh, this is about cryptocurrency, and I don’t need to get involved in that.’ Actually, it’s very important, because it’s also about the sovereignty of currency,” she added.

Bryan Pascoe, chief executive of the International Capital Markets Association (ICMA), said the ECB’s posture is deliberately defensive.

“The European priorities have been to take a more defensive approach and look at things more related to central bank-run processes than we’ve seen in the US,” he said. “That doesn’t always provide the full-scale accessibility you would see from a private sector stablecoin.”

Europe lacks a single safe asset deep and liquid enough to back a euro-denominated stablecoin at scale. Even Germany, France and Italy have relatively illiquid bond markets compared to the US. That is why the ECB is focused on delivering a central bank digital currency (CBDC) as its digital solution.

Jannah Patchay, executive-in-residence and board member at Global Digital Finance (GDF), said the UK’s current legislation leaves stablecoin holders exposed.

“There is a black hole around the rights that coin holders have,” she said. “There are no provisions as to how, if I’m holding a US dollar-denominated stablecoin, I am actually meaningfully able to redeem them.”

The Financial Conduct Authority (FCA) is now building a competitive framework for locally issued stablecoins. Cameron founded the first All-Party Parliamentary Group (APPG) on digital assets in 2020, when there had been no debates on the subject in the House of Commons. She said parliamentary engagement has since “tapered off” and must urgently be revived.

Synthetic CBDC emerges

The panel took place at the Financial Times Digital Assets Summit in London on May 13. It was moderated by Jill Shah, US trading and crypto correspondent at the Financial Times. The discussion examined whether stablecoins promote financial inclusion or risk entrenching dollar dominance globally.

One idea that drew notable interest was a CBDC-stablecoin hybrid, raised by an audience member from a US bank stablecoin issuer. He proposed that the Federal Reserve allow banks to earmark a sub-account of a Fed master account for stablecoin reserves, removing capital requirements and shifting them to the Treasury’s balance sheet.

Patchay said the structure would entirely eliminate the US Treasury as an intermediary in US Treasury issuance.

“It’s almost like a synthetic CBDC type structure,” she said. “But ultimately, it’s the coin holder who is lending to the government via you as an intermediary, just without the issuance of any debt at any point. From a regulatory perspective, it’s super interesting.”

Cameron said a CBDC-stablecoin hybrid had been widely discussed at the Consensus Miami 2026 conference in early May as a likely direction for the longer-term future.

On wholesale markets, Pascoe was candid about how far the industry has to go. Digital securities have undergone extensive pilots but have yet to achieve scale, largely because cash settlement has failed to keep pace with the digitalization of securities.

“We still see stablecoins principally as an access tool for crypto asset trading,” he said. “They haven’t necessarily manifested themselves on a consistent basis in terms of payments, savings or remittances, which is the next obvious use case.”

Only the US dollar market has the depth to support stablecoins as collateral at scale. Tether is already the 17th largest buyer of US Treasuries, arriving faster than most expected, yet its presence has already moved yields.

Pascoe also cautioned that the label misleads retail investors, who may assume a stablecoin guarantees a one-to-one dollar peg, which it does not. He said the inherent risk needs to be communicated more clearly.

He said short-dated government bonds remain the most important collateral for stablecoins because they are safe, provide a reasonable yield to issuers, and can be liquidated quickly to meet redemption demands.

The Bank of England’s ongoing stablecoin consultation and the UK Debt Management Office’s work on expanding Treasury bill issuance are likely connected for precisely this reason.

Africa’s payment detour

Last year’s GENIUS Act enshrined an explicit US policy intent to entrench the dollar's global use through stablecoins.

Patchay said this poses a direct challenge to monetary sovereignty wherever the dollar is not the native currency.

“The combination of the explicit policy intent from the US government that stablecoins should be another means of entrenching dollar use in the financial system, along with the ready availability of dollar-denominated stablecoins, poses challenges to monetary and economic sovereignty,” she said.

“If you clamp down entirely on dollar stablecoins, people will just use them in ways that are not transparent to the regulator. Regulators and central banks will lose sight of those flows and lose all control,” she added.

Nigeria and China have both attempted restrictions, pushing usage underground rather than eliminating it. Nigeria’s e-Naira CBDC also illustrates a key lesson: tokenizing a currency does not work if the underlying economic fundamentals are weak.

The challenge is sharpest in Africa. Colonial-era correspondent banking routes payments between African countries through European and American financial hubs.

“If you are in Senegal in West Africa and you want to make a payment to a supplier in Kenya in East Africa, your payment will be routed from Dakar to Paris to New York for the dollar leg, back over to London and then down to Nairobi,” Patchay said.

“There will be multiple correspondent banks involved at each leg of the journey, and therefore multiple layers of cost and inefficiency. Some of these payments can take up to a week or more,” she added.

African central banks are now looking to leapfrog this system with stablecoins and digital money, just as the continent once leapfrogged traditional banking with mobile money. This year, GDF launched an Africa-specific chapter on stablecoin regulation at the 3i Africa Summit.

Cameron added a psychological dimension that policymakers have largely overlooked.

She said that as money becomes digital, citizens’ emotional connection to their national currency weakens. She said governments must urgently build frameworks to protect monetary sovereignty before that attachment fades entirely.

Stablecoin issuers on both sides of the Atlantic will not be allowed to pay interest to holders for at least the next five years, cementing their role as payment tools rather than savings instruments and allowing yields to flow quietly but lucratively to the issuers themselves.

The regulatory walls are now firmly in place. Both the EU’s Markets in Crypto-Assets Regulation (MiCA) and the US GENIUS Act explicitly prohibit stablecoin issuers from passing interest to holders. As the market matures, the question of who captures that yield is becoming impossible to ignore.

“If we could, we would love to pay interest, but we can’t,” said Simon Seiter, chief financial and product officer of AllUnity, Europe’s first regulated stablecoin issuer. “With both MiCA and the GENIUS Act basically having that framed, I think it’s impossible in the next five years to change it.”

He said the yield flows to the issuers, just as banks earn interest without passing it all on to depositors.

“That’s fine, as long as you provide value,” he said. “For payment mechanisms, people pay a lot, especially on the merchant side. If we only take the yield, they say, ‘Please take my money. I will use you just to save the fees on the payment side.’”

How lucrative that yield can be was illustrated bluntly by Seiter.

In 2024, Tether, the world’s largest stablecoin issuer with around 150 employees, generated approximately $14 billion in profit, roughly matching Goldman Sachs, which employs 65,000 people, in the same year.

Elliot Hentov, chief macro policy strategist at State Street Global Advisors, said the number-one issue the American Banking Association raised with US government officials in 2025 was stablecoin yields and rewards.

“The path is hard. If I’m a regulator, it just doesn’t jive with my existing system, unless I completely redesign everything else,” Hentov said. “There may be back doors, reward programs and so forth. But I still struggle to see that happening in the near term.”

Seiter said regulated issuers such as AllUnity cannot compete with unregulated rivals on decentralized finance (DeFi) yields because their reserve structures are closer to those of money market funds. For merchants, the savings on payment fees alone are sufficient incentive to adopt stablecoins even without yield.

Trump’s debt lifeline

The discussion took place at the Digital Assets Forum 2026, organized by the European Blockchain Convention in London. It was moderated by Mads Clemmensen, digital assets lead at Danske Bank.

Speakers from JPMorgan, State Street Global Advisors, Zero Hash, and Innovating Capital joined Seiter to examine whether stablecoins are becoming competitors to money markets.

The numbers are stark. Total stablecoin supply currently stands at roughly $170 billion, of which euro-denominated stablecoins account for only around $600 million. In traditional currency markets, the euro accounts for 20 to 25% of US dollar trading volume.

Seiter said the gap between those two figures measures how aggressively the stablecoin market is driving dollarization.

“Why is Donald Trump regulating stablecoins?” he said. “The GENIUS Act requires issuers to invest in short-term US Treasury bills. Tether was already among the top 50 buyers of US debt in 2024, bigger than the state of Germany. A new buyer group comes into a saturated market and buys your debt, and it becomes significantly less expensive.”

“Stablecoins solve Donald Trump’s short-term liquidity problem, because they make his refinancing cheaper,” he added. “That is geopolitical, and it is not a development that, from a European perspective, has a beautiful future.”

He said the systemic risks embedded in that arrangement are considerable. Tether mixes Bitcoin into its reserves alongside Treasury bills, creating vulnerability during crypto bear markets. If Tether were to fail, ordinary users in emerging markets such as Argentina, who hold USDT (Tether’s dollar-pegged stablecoin) as a store of value, would bear the consequences.

Tether’s newly announced regulated US stablecoin, USAT, will not take effect until 2028, leaving USDT itself unregulated and meaning systemic risk remains largely intact. Tether already holds the 12th-largest position in US government debt.

Rens de Groot, chief commercial officer for Europe at Zero Hash, which provides blockchain infrastructure to banks and brokerages, said the dollar dominance of stablecoins reflects a broader shift in the world order.

“The dollarization of the world gives the Trump administration just another tool of creating financial influence throughout the world,” he said. “Here in the EU, we should be more than aware of this. The supply of stablecoins should be a reflection of real-world economic size.”

Hentov said the risk is most acute in non-dollar economies, where the appeal of holding a dollar-denominated asset is already strong and local banking systems face erosion as a result.

“In non-dollar areas of the world, holding the dollar has a variety of benefits and attractions. You are going to get a hollowing out of the banking sectors, or an erosion of the banking base,” he said. “These are not great options.”

He said the UK is actively reconsidering its stablecoin regulatory framework. The ultimate prize is a non-dollar settlement currency: global equity and bond markets each stand at roughly $100 trillion, and tokenizing even a fraction of those assets would require one.

Emma Lovett, markets digital assets lead at JP Morgan, said the arrival of a major asset manager’s own stablecoin raised the question of whether the buy side would begin transforming its payment infrastructure over the next two to three years.

JP Morgan launched its own deposit token, JPM Coin, in November last year, viewing it primarily as an institutional payment instrument.

Fruit rotting at port

Beyond regulation, the industry faces a structural challenge: a proliferation of competing issuances that risks fragmenting liquidity and deterring corporate adoption.

De Groot said he recently addressed a group of corporate treasurers in Amsterdam for whom fragmentation was a source of genuine alarm.

“We are all crypto natives here. We understand there’s more than one euro stablecoin and more than one USD stablecoin. But those treasurers found it really scary,” he said. “Their consensus was that they hope it will be so interoperable that in the end they will not care, or will not even see, which stablecoin is being used in the back end.”

Seiter drew a parallel with the existing banking system.

“In the existing banking rails, you have a lot of different private money, and you don’t notice. If I send you a bank wire, I have maybe Deutsche Bank money, you have maybe Citibank money. It’s all different money by different issuers, and nobody notices because it’s already integrated,” he said. “For stablecoins, it will be the same in the future.”

De Groot described a customer whose container of tropical fruit was rotting in an African port, waiting for a SWIFT payment that had not arrived.

“He said to me: I’m not interested in blockchain technology. I’m interested in my container of tropical fruit. How are you going to solve that for me?” he said.

“When your payment goes through many correspondent banks, in the end there’s no way of telling which amount will actually arrive at the African port,” he said. “With stablecoins, clients can just log into a public blockchain viewer and see where the cash is. It cannot be anywhere else.”

Anthony Georgiades, founder and general partner of Innovating Capital, a New York-based technology infrastructure fund, said the most meaningful venture capital is now flowing not into new stablecoin issuances but into the invisible infrastructure beneath them.

“A lot of the meaningful investment right now is really going into that invisible layer of backing infrastructure, things like orchestration: who decides in an autonomous fashion when to use stablecoins, and how do I bridge stablecoins across different ecosystems?” he said.

He said the industry would reach true product-market fit only when stablecoins are so embedded that people stop discussing them altogether. He said a core focus of Innovating Capital’s portfolio is to abstract that complexity away so that banks can adopt stablecoins without needing to understand the underlying blockchain architecture.

Europe’s fragmented regulatory landscape is emerging as the defining obstacle to large-scale bond tokenization, while the United States moves ahead with sweeping regulatory clearance, leaving the continent locked in a legislative process that could delay the development of meaningful market infrastructure until the end of the decade.

The contrast could hardly be sharper. The contrast could hardly be sharper. In December 2025, the US Securities and Exchange Commission (SEC) issued a no-action letter to the DTCC [Depository Trust and Clearing Corporation], clearing the way for a national-scale tokenization pilot program. Europe's equivalent framework is not expected to be operational until 2029–2030.

“We have 27 national regimes in Europe. There is no real harmonization. Yes, we do have MiCA [Markets in Crypto-Assets Regulation], but honestly, still 27 regimes, 27 national regulators, 27 bunches of problems for everyone,” said Jan Klesla, Chief Designer and Head of International Relations at DEUSS, which is building European public DLT infrastructure for small and medium-sized enterprise (SME) bond markets.

Mark Kepeneghian, Vice President of ADAN, the European digital asset industry association, spelled out the regulatory disparity. The SEC's no-action letter to DTCC, which holds a monopoly as the country's central securities depository, enabled tokenization to proceed at scale with minimal regulatory friction. In Europe, the DLT [distributed ledger technology] Pilot Regime 2 was reintroduced into the legislative process in December, but will not be live until 2029–2030.

Michael Ashby, chief executive of AlgoQuant, a multi-strategy hedge fund, said he does not see the situation as a zero-sum competition. Europe has deep financial history, and the UAE is already issuing tokenized securities and real estate bonds ahead of both the US and Europe. He said once Europe's 27 regulators eventually align, it will help the entire global ecosystem develop.

Klesla agreed that the UAE’s regulatory approach, which draws heavily on European frameworks, could serve as a basis for broader international coordination. Europe can still lead if it is willing to work openly with the UK, the US and other jurisdictions rather than treating harmonization as an internal matter.

Demand already institutional

The panel was convened at Digital Assets Forum 2026 in London, organized by the European Blockchain Convention. Kepeneghian moderated a discussion on the practical mechanics, demand dynamics and liquidity challenges of tokenized bond markets.

He was joined by Klesla; Ashby; Jessica Hakizimana, Alternative Investments Program Manager at Belfius Private, the second-largest Belgian private bank; and Lorenzo Savi, Board Member of Falcon Investment Management, a crypto and digital asset manager.

On the demand side, Hakizimana made the most comprehensive case for tokenized bonds. Research from the University of Bayreuth in Germany found that tokenization can automate more than 2,000 tasks in the bond issuance process, saving roughly 1,000 person-hours per issuance. What previously took 12 weeks can now be compressed significantly, with bookkeeping periods cut by more than 50%.

A separate McKinsey study from 2024 found that end-to-end tokenized bond lifecycle efficiencies of at least 40% are achievable.

“Tokenization is not just a new narrative. It’s not an asset class. It’s a new distribution strategy, and the real impact comes when you manage to connect those tokenized assets and tokenized bonds to powerful channels,” she said.

She pointed to a generational wealth transfer as a major structural driver. UBS estimates that approximately $84 trillion will pass to the next generation globally by 2048, with $6 trillion of that among billionaires alone by 2040. The bulk of this transfer is happening within the next 10 years.

Next-generation beneficiaries in Europe already allocate on average more than 12% of their portfolios to crypto and digital assets. A recent EY report found that close to 70% of high-net-worth individuals plan to allocate 5% or more of their portfolios to tokenized assets.

“We have crossed the threshold from hype to institutional strategy. When you see BlackRock saying that they are expanding their tokenized assets and tokenized offering, it’s a game changer,” she said.

She also cited the NASDAQ, the New York Stock Exchange and the CFTC [Commodity Futures Trading Commission], which issued guidance in 2025 confirming tokenized assets can be used as collateral. SEC Chairman Paul Atkins has added further weight to the institutional shift.

Despite the strong case for demand, Ashby said the secondary market for tokenized bonds remains underdeveloped.

“Tokenization has been around for a while now, and there is really no demand function. There is a lot of supply of paper, but the reality is that the demand function has been quite low,” he said.

The problem is largely one of distribution. Crypto markets developed through retail channels and technologists rather than through the institutional frameworks that govern bond markets. Morgan Stanley recently rehired a digital assets team, though this happened only weeks before the forum and will take time to translate into active market-making.

On the decentralized side, HyperLiquid, a DeFi [decentralized finance] perpetuals platform, has emerged as the dominant player over the past six months.

During silver's surge to an all-time high in late January 2026, more than 10% of the silver supply was traded in tokenized form on the platform. Such volumes will never attract institutional participation due to KYC [Know Your Client] and AML [anti-money laundering] requirements, but they demonstrate genuine underlying demand for tokenized traditional assets.

“Tokenization is the golden ticket that actually brings institutional capital into crypto. We do think there is a huge market, first for bonds, but then for structured products. If you look at gold, there is a huge structured-product market for gold and oil. There is nothing for Bitcoin in relative terms,” Ashby said.

He estimated the market is 12–18 months away from meaningful institutional liquidity, contingent on both regulatory progress and technology upgrades. A significant technical barrier remains legacy back-office infrastructure: token quantities can extend to 20–30 decimal places, but most traditional systems cap at eight, making it impossible to book the trades.

With most large institutions carrying 20–30 years of technology debt, rebuilding systems to handle tokenized instruments will take considerable time.

Italy leads early issuances

DEUSS is building its infrastructure on top of a ledger developed by the European Commission, giving conventional institutions a regulated and familiar anchor point.

“To build a market that will help finance the real economy, you need infrastructure that offers two things: all the benefits of decentralized technology and the blockchain space, but also it has to be trusted — the trust of traditional finance and the traditional real economy,” Klesla said.

“In five to 10 years, we will be able to serve a usual entrepreneur in Norway, in Greece, in Spain, in Scotland with the whole spectrum of financial products based on DLT infrastructures,” he said.

DEUSS emphasizes full interoperability across public and private ledgers, financial intermediaries and distributors, while issuing standardized bonds that conform to existing legal frameworks. Reducing intermediaries is a long-term goal: SMEs seeking financing will continue to work through banks, brokers and investment firms for now.

Real-world evidence of early adoption is emerging from Italy. Lorenzo Savi, Board Member of Falcon Investment Management, noted that the Italian SMEs leading tokenized bond issuances are predominantly technology and artificial intelligence (AI) companies.

Metrica, an AI and data solutions provider, issued a €1.8 million tokenized bond with Banca di Sabina, a regional Italian lender. UniCredit and Cassa Depositi e Prestiti (CDP) underwrote a separate €5 million tokenized bond issued by E4 Computer Engineering, an engineering and AI firm.

“The SMEs that tend to issue tokenized bonds in Italy are either AI-related or AI-driven. They already have technology and digital in their DNA,” Savi said.

At the institutional level, Falcon Finance is in advanced discussions with fund managers interested in launching what could be the first portfolio composed entirely of tokenized bonds.

“At Falcon, we are in talks with a couple of fund managers that want to launch a portfolio dedicated only to tokenized bonds,” he said.

Family offices and private banks are well positioned as early adopters, Hakizimana said, given their clients’ appetite for new asset classes, lower minimum tickets and greater transparency. Trust and education are the final critical bottlenecks.

“Education is not just about educating the investors, but also the regulators and the financial advisors,” she said. “Trust is verifiable on blockchain.”

Savi drew a parallel to other disruptive technologies, noting that adoption curves tend to be slow at first before accelerating sharply.

"As with any new and disruptive phenomenon, at the beginning you need to build up an ecosystem. It progresses slowly, and then at some point there will be exponential growth, perhaps in two to three years," he said.

With infrastructure consolidating, institutional interest building and early issuances proving the model, panelists agreed that the remaining variables are regulatory alignment, technology modernization and time.

Companies seeking to register with the U.S. Securities and Exchange Commission (SEC) during the Biden administration were subjected to enforcement actions simply for asking questions, a practice that one senior official now says was legally indefensible and could not, in good conscience, be continued.

The SEC’s crypto task force, created on January 21, 2025, under new chair Paul Atkins, has since overseen one of the sharpest reversals in U.S. regulatory history. The official at its center says the rollback was not politically motivated, but legally necessary.

“I was part of meetings where people came in and said, we want to register, we want a pathway, and then an enforcement action is brought against them three months later, just for coming in and talking and asking questions,” said Sumeera Younis, Chief of Operations of the SEC’s Crypto Task Force.

“So not only have they not been given a pathway, now they’re being punished for activity that they don’t even have a way to properly conduct,” she said. “If the premise of all of those enforcement actions was something that was foundationally wrong, then it’s not right to keep pursuing those.”

The rollback has been sweeping. The SEC dropped or closed at least 12 crypto enforcement cases from early 2025, including major actions against Binance, Coinbase and Kraken.

The Coinbase case was dismissed on February 27, 2025, the same day SEC Commissioner Caroline Crenshaw issued a public dissent, coining the phrase “regulation by non-enforcement” and warning it ignored 80 years of established securities law.

Younis said the previous approach, widely described as regulation by enforcement, saw companies punished simply for seeking a registration pathway, before any viable legal route had been established for them to follow.

She pushed back on characterizations that the task force was staffed by crypto enthusiasts. Much of the press coverage of the enforcement rollback has been inaccurate, she said, partly because SEC staff are not permitted to speak to the press.

"There's this mischaracterization that the crypto task force is a bunch of crypto bros, but there's not a single ‘Patagonia vest’ [a branded fleece gilet popular among finance and tech bros] in our office," she said.

She has been at the SEC for a decade, previously working on fund registration before joining Commissioner Hester Peirce’s office, where she helped develop the conceptual framework for the task force ahead of the change in administration. Before the SEC, she worked in-house in the financial industry.

Howey still the law

Younis was speaking at the Financial Times Digital Assets Summit in London on May 13 in a keynote interview moderated by Jill Shah, the Financial Times‘ US trading and crypto correspondent. The discussion covered the SEC’s rebuilding of its relationship with the crypto industry, jurisdictional questions, investor protection and the divergence between US and UK regulatory cultures.

She applied three core criteria when recruiting task force members: at least five years at the SEC, an independently developed interest in crypto even when the agency was hostile to the asset class, and intellectual humility.

“I wanted people who, just out of their own curiosity, in an environment that was actually pretty hostile to crypto, were asking questions,” she said. “They wrote memos on their own accord. They’re already thinking about this, just out of intellectual curiosity.”

The team comprises lawyers specializing in distinct areas of U.S. securities law, including the Securities Act of 1933, the Securities Exchange Act of 1934, and the Investment Company Act of 1940.

The task force has held hundreds of industry meetings, likely more than 1,000 in total, and issued a public request for information comprising 48 foundational questions. Submissions were made public, allowing respondents to engage with each other’s arguments.

She said she had no idea how effective policy work could be done without engaging with industry.

“Your industry experience grows stale pretty quickly once you’re outside of it,” she said.

The Howey test is a four-prong framework established by a 1940s U.S. Supreme Court case involving an orange grove. It determines whether an asset constitutes an investment contract and therefore a security.

On March 17, 2026, the SEC and the Commodity Futures Trading Commission (CFTC) issued a joint statement providing interpretive guidance on five categories of crypto assets, assessing each against those four prongs: an investment of money, in a common enterprise, with a reasonable expectation of profits, derived from the efforts of others.

“The question that you triple underline is: what is a security? Because that led to a lot of the cases that were brought, and also a lot of the uncertainty in the marketplace,” she said. “If you don’t know what you’re dealing with, a security or not, you don’t know if you’re in our jurisdiction, the CFTC’s jurisdiction, or out of both.”

She said every one of the five categories in the March 17 statement was assessed against the Howey test, and those that failed on even one prong fell outside the SEC’s jurisdiction. The SEC was not trying to give crypto assets a free pass. Those categories simply were not securities, and the agency was not the regulator of everything in the world.

CFTC chair Brian Quintenz previously served as chief counsel of the SEC’s crypto task force, giving the two regulators an unusually close working relationship as they move in lockstep on crypto policy.

Blank page on tokenization

The task force has lost sleep over striking the right balance between investor protection and the creation of a regulatory environment where businesses can thrive, Younis said.

The roughly dozen members share the view that departures from previous rules must be deliberate and legally grounded, not a signal that businesses can act without constraint.

“I want an environment in which builders can build, but not at the expense of regular people who trust these markets to invest in,” she said. “None of us want investors’ lost assets on our conscience. None of us want a financial market event on our conscience.”

The task force’s output has been slow because of the rigor of its process. The team spends dozens of hours per week on policy work beyond its regular duties, and its internal diversity spans lawyers from different statutory backgrounds and professional experience.

“Some of this stuff is taking so long to get out the door because we spend dozens of hours every week just doing policy work, sitting there, really interrogating what we’re doing, and challenging each other,” she said.

On the differences between US and UK regulatory approaches, the two jurisdictions reflect fundamentally different national risk appetites.

She was largely unfamiliar with the Financial Conduct Authority (FCA) and His Majesty’s Treasury (HMT) until around a year before the interview. As an immigrant to the US, she said American social mobility and the openness of its capital markets stem in part from a higher tolerance for risk.

She acknowledged recent UK signals from officials including Chancellor Rachel Reeves, indicating a desire to rebalance toward growth.

“You do have to take on a little bit more risk for growth, but that doesn’t need to be reckless risk, and it doesn’t need to be something that allows investors to become vulnerable to exploitation,” she said.

She identified tokenization as the area she is most excited about, describing it as a space where virtually every jurisdiction is starting from scratch with its regulatory framework.

“Most jurisdictions have a blank page in front of them in terms of the regulatory framework around tokenization,” she said. “I think it’s an opportunity for the two greatest financial markets in the world to work together.”

Younis confirmed she is relocating to the UK and expressed hope of contributing directly to US-UK regulatory collaboration on tokenization, bringing some of America’s risk appetite to this side of the Atlantic.