Clive Foley

His view of AI is unapologetically optimistic

November 2, 2026

Clive Foley runs engineering teams inside one of the most distinctive units in global insurance broking. His view of AI is unapologetically optimistic, grounded in delivery, and built on a premise: the technology is now the easy part.

Most conversations about AI in large enterprises eventually arrive at the same place: headcount, efficiency targets and the quiet arithmetic of simply doing the same thing with fewer people. Clive Foley's conversation goes somewhere else entirely.

Foley is a Director of Engineering in the Innovation Centre at Marsh, the global professional services firm formerly known as Marsh McLennan, which adopted its new single brand at the start of 2026. Around three months into the role when we spoke, he described himself as having landed somewhere unusual: a part of a large, regulated, publicly listed business where every AI story he works on is a growth story.

"Depending on where you land on the spectrum with AI these days, you're either like, it's going to change the world, or it's complete doom and gloom," he says. "With this role, I found a nice middle ground leaning towards positivity, because all the stories are around growth and enabling our colleagues. It's fun from a technology perspective, but you can see how it's going to impact the bottom line. You can see how it's going to help colleagues. So there's win-win everywhere."

That framing matters. In a market where much of the AI narrative is about cost, Foley's remit is to find new value, move quickly and bring the rest of the organisation with him. The Innovation Centre exists to find the edges of what is possible, push through them and then unblock the path for everyone else. It carries a mandate from across the group to do exactly that.

"Our job is to move fast, find all the edges, push through them a little bit, and then get people to come with us."

— Clive Foley, Director of Engineering, Innovation Centre, Marsh

"Our job is to move fast, find all the edges, push through them a little bit, and then get people to come with us," he explains. "Where you find a block, go find somebody who will help you unblock it. And those people are empowered to help us."

It also gives his work something that much AI investment still lacks: a direct, legible line to commercial outcomes. "We can directly point out the improvements we are making generate dollars," he says. "When you're just talking technology, and some of the more abstract terms like productivity, well, how do you measure that? Whereas when you say we did this, that went faster, therefore more dollars appeared over there, that becomes very easy to tie to an outcome that everybody can understand."

From Purist to Partner

Foley has been writing code since he was very young, and for most of his twenty-plus year career his professional identity has been bound up in the craft of it. How code was written, how it was structured, how elegant it was. He describes himself, without much hesitation, as having been a bit of a purist.

The last eighteen months have changed that. Not by replacing his identity as a builder, but by changing what building means.

"I've become more and more reliant on a coding agent to do things, and I'm becoming more trusting of it," he says. Through carefully maintained readme and markdown instruction files, he has taught agents not only the swim lanes they should work within, but how he likes code to look. "I've gotten to a point where I have an almost really cheap version of myself that is able to solve problems with my direction, and will create output that I would be happy with. If you'd said that to me three years ago, I'd have said, yeah, we're going to get there. But in three to six months? Not a hope."

The most significant shift is not in the code itself but in how he thinks about time. Two decades of instinct have given him a reliable internal clock for estimating work: this is a week, that is a fortnight, that is a quarter. He now deliberately overrides it. "I'm challenging myself going, no, not anymore. You're probably off by a factor of two, possibly a factor of three, depending on the problem set," he says. He has also found himself working confidently in areas that were never his natural strength, describing the problem and the outcome he wants and letting the agent carry more of the implementation.

His advice to others is borrowed from the cinema. In a scene from Inception, one character is holding off attackers with a handgun when another arrives, produces something considerably larger, and tells him he mustn't be afraid to dream a little bigger. Foley has turned the line into something of a personal mantra.

"That's what I tell everybody. You mustn't be afraid to dream a little bigger, because this thing is not just a five or ten percent improvement for the right people," he says. "This is a force multiplier that is just insane, in my opinion."

The Commercially Orientated Builder

When Foley talks about "the right people", he is precise about what he means. It is not the most technically gifted engineer in the room. It is the person who understands how the business actually makes money.

"It's never been a better time to be a technologist who is business focused," he says. "If you're the kind of person who understands the business, understands the KPIs, what the levers are that move things, you can probably collapse three or four roles into one."

Put a handful of those people together with a business counterpart who thinks the same way, people senior enough to reason about systems rather than features, and the effect is dramatic. "Those people are dangerous," he says. "And I mean that in a good way. They're very dangerous."

He is candid that this is exactly what his own leadership noticed when he joined. The feedback from senior colleagues was not about technical capability. It was about how quickly he had absorbed the business. A background in reinsurance helped, but the real skill was in tracing cause and effect: this process creates a bottleneck for this many people, removing it releases pressure downstream, and that flows through to revenue in specific ways.

"After that, together with an AI coding agent, the technology is easy," he says. "Where should you apply your focus and why? What does success look like? And then how do we get there? There used to be this huge building phase afterwards. But the time, effort and cost of the building phase is collapsing towards zero."

"The time, effort and cost of the building phase is collapsing towards zero."

That time reduction is reshaping roles well beyond engineering. Foley recalls working with an exceptional designer in a previous role, someone with genuine taste who had lived in Figma for most of his career. Once a few features had established the product's visual language, the engineers could feed that design system to their agents and generate a strong first pass for the next feature, typically eighty-five to ninety-five percent of the way there. The engineers began to run ahead.

"It's not that their talent is no longer needed. Not at all. Their talent is more than needed, particularly in design," he says. "It's just that the speed and pace previously required is no longer the pace needed to run this race. And when I say they, I mean everybody."

He frames the choice facing every craftsperson as a trade-off rather than a verdict. Someone working at a faster pace will often deliver seventy, eighty or ninety percent of what a slower, more deliberate colleague would. "The question is, are you comfortable sacrificing time for the delta, which is getting shorter? There's no right answer there. It's a risk decision."

Asked which role ultimately stands out as functions converge, Foley does not hesitate. It is the commercially orientated builder, and he defines builder broadly. "If somebody can vibe code something up and they're non- technical and they can get there, then to me, they're a builder in this modern world," he says. "If you can make money, that's the ultimate way to express your value to an organisation or to the market."

The Temufication of Software

Foley has coined a phrase for what he believes happens next to a large swathe of the software economy. He calls it the Temufication of software and businesses, a nod to the Chinese marketplace that wins on price.

The logic runs like this. Take any established classifieds or listings business. It has a long history, a large engineering team and a substantial payroll to support. Its most valuable asset is usually its brand, not anything that cannot be copied. "They don't have the formula for Coca-Cola," he says. "A couple of ambitious young people with a coding agent can replicate the business, but charge one tenth of the price, because they don't bring the overheads."

If it costs a euro to list on the incumbent, the challenger charges twenty-five cents. Repeat that across every directory and listings model, and incumbents face entrants operating at a price point they structurally cannot match. "If it's you and two buddies and a couple of coding agents, and you can generate a couple of hundred grand and cover your own salaries, how much fun is that?" he asks.

He does not predict overnight collapse. He sees a pattern closer to fintech's assault on banking a decade ago: challengers nipping at the edges of bundled business models, each taking a slice. "Will businesses collapse overnight? No. But they'll probably end up with more challengers in a shorter time frame than they ever expected."

The more interesting idea is what this does to the economics of entrepreneurship itself. For a generation, founders have been told that small markets are not worth the effort, because the cost and risk of entry were too high relative to the return. Foley thinks that calculus has inverted. "You don't need one business to generate four or five million a year. That's actually quite risky," he says. "What you probably need is half a dozen businesses that all generate three hundred grand. If any one of those industries goes south, you still have your portfolio. So, it's actually better to de-risk things in that regard."

For early-career graduates facing a tight junior hiring market, he believes the moment is extraordinary. "It's like taking the world of business like a snow globe and just shaking it. Things are still floating in the mixture, and nothing has settled yet."

The corollary for established companies is a sharper question about where defensibility really sits. Foley's own answer, as we will see, points squarely at expertise, judgement and relationships, rather than code.

A Startup Inside a Corporate

The Innovation Centre at Marsh has existed for a number of years, but its AI focus has sharpened considerably over roughly the last two. It began with narrow technical questions, can you do this specific thing, and after a run of successes it has moved on to something more ambitious: reimagining entire business capabilities.

The operating model is disciplined and deliberately incremental. An idea arrives from a business within the group. A product lead and a technical lead sit down with the subject matter experts who live with the problem and spend a few days working out what the problem really is and what success would look like. The output is a thirty-day experiment designed to validate or invalidate the team's assumptions as quickly as possible.

"It's very much a dial-up motion," Foley explains. If the early work shows promise, a small team of two or three more people joins to build a working prototype that demonstrably changes something in the subject matter expert's day. Then comes the decision point: do the results justify scaling?

"You're not committing millions up front. You're committing, at max, thirty days of three to four people."

"You're not committing millions up front. You're committing, at max, thirty days of three to four people," he says. "That's large enough that you can test something, but also small enough that if it doesn't work out, well, we didn't bankrupt the company."

There is a pleasing irony in the approach, and Foley is quick to point it out. Insurance is a risk business, and the Innovation Centre essentially applies the firm's own discipline to itself. Small, controlled bets. Measured exposure. Scale what works. "They're using their own expertise to move to the next level," he says.

The early results speak for themselves. Of the five ideas that came to his teams when he joined, two have already moved into product and are being actively built out. One remains in its design sprint, and two more have completed their sprints and are under review.

The teams themselves are lean: full stack engineers, a product owner, and direct, regular access to the business. Foley singles out that last ingredient as the real secret of the model. Subject matter experts across the firm give their time generously, not once but every week, and often several times a week.

"If you need insight on a quirky insurance concept that you don't know, there are people there who are more than happy to walk you through the nuance," he says. "People are enthusiastic. They want to help. I'm at this twenty-odd years, and I can probably tell you three or four places where I've had a similar feeling. They've nailed the culture."

For an engineer, he adds, the domain itself is part of the appeal. Business insurance in particular layers policies, coverages, limits and endorsements across multiple jurisdictions and regulatory regimes. "I've described it to colleagues as a Rubik's Cube," he says. "Just when you think you've got one side done, you flip it over and you go, oh, I missed a few squares here. If you like puzzles, insurance is a fantastic place to go."

The overall experience, in his words, is like running a startup inside a corporate. "Running a startup, you're like, if this doesn't work out, we don't eat next week. Here, it's a safe place where you can play. We have to produce something of course, but you have all the resources of a corporate there, so if there's any spark at all, there's fuel to push it to the next level. That's what startups don't always have."

A High-Class Problem

Every innovation function eventually meets the same question: what happens when the experiment works? Foley is living that question in real time.

One of his products began with a pilot group in the double digits. It is now moving to an audience in the hundreds, and will eventually reach the thousands, all internal. Along the way, internal IT colleagues have worked alongside the Innovation Centre on infrastructure, CI/CD, security, data residency and privacy, so that the R&D work can succeed safely inside a regulated environment. Now that the signals are clearly positive, demand from colleagues has accelerated. "Our internal colleagues are screaming at us, we want that yesterday," he says.

The natural model is to hand the product over to the group's internal teams. But Foley is honest that it is not that simple, and that the right answer is still being worked out. "To get to where my team is today, we had to learn so much about insurance," he explains. "If you hand it over one hundred percent, from an accounting perspective the internal team now owns it, they own the costs. But you're going to drop so much domain knowledge on the floor. There's no perfect answer there."

A blended team, where innovation and internal delivery continue to work side by side, is one option on the table. What strikes him most is the tone of the debate. "There is no one group saying we should definitely have it or we definitely shouldn't. Everybody's weighing up the pros and cons. It's collaborative, and I'm sure we'll arrive at the right answer because everybody's coming at it from the right place."

It is a challenge that will feel familiar to any organisation scaling AI beyond the lab. The knowledge that makes an experiment succeed is rarely written down, and it lives with the people who built it. Foley borrows a phrase from a former colleague to describe the situation, which one his new teams now tease him for repeating. "This is a high-class problem to have," he says. "These are the types of problems that you want to have."

Where the Model Ends and Determinism Begins

Ask Foley what excites him most technically and the answer is specific to his world. Insurance runs on documents. Contracts pass back and forth, policies are written in dense natural language, and very little of it arrives as structured data. The steady improvement in how models understand language is, for his teams, the single most important development of all.

"Everything is documents," he says. "So the fact that it can understand the documents more and more, that is our current bottleneck. And the whole point is it collapses time and it collapses cost. If you can do that, you can massively increase volume. That's what I mean by the growth opportunity."

He is equally clear about the limits, and this is where the most practically useful insight in the conversation sits. The first proof of concept almost everyone runs is to put a document into a large language model and see what comes back. On a simple policy, the results are impressively accurate. Then reality intervenes.

"The second you give it any level of complexity or length, it starts to break down very, very fast. Or you don't get repeatable results," he says. "So you're trying to figure out where the line is at which the LLM gives maximum value relative to the probability of it being right. And then where you, as a human, have to come in with something that's deterministic after that."

That boundary, between probabilistic intelligence and deterministic engineering, is where he believes the real craft of enterprise AI now lives. It is not glamorous, and it rarely features in product launches, but it is what separates a convincing demo from something a regulated business can rely on.

Foley's own moment of conviction came from a similar place. Experimenting with a personal side project, an agent that monitors utility tariffs on his behalf, he fed a model his electricity bills and asked it to tell him something he did not already know. It noticed low daytime consumption and heavy usage in cheap overnight windows, and concluded that the house almost certainly had solar panels and a battery. It was right. "It did a hop, skip and a jump, and where it landed was very accurate," he says. "As a user, that was the magic moment."

He draws a broader lesson from it about where value is moving. For two decades, much of enterprise technology was about integration: connecting systems and moving data between them. Models are making that almost trivial. "Integrations are just nothing now," he says. "What's pushing value up is coordination, intelligence, almost having an opinion on something. That's where all the real opportunities are."

And that, he notes, is precisely what Marsh does well. Clients seeking business insurance typically need multiple lines of coverage across several jurisdictions, and much of the value lies in advice: understanding a client's situation, understanding the market, and guiding them through it. "Understanding your policy is step one," he says. "Now the LLM is helping to get that out of the way so we can get to the advisory stage quicker. The insurance industry is about relationships. It's about understanding what people need, not only today, but next year and the year after, and going on that journey with them."

In other words, AI is not replacing the moat. It is clearing the path to it.

Two Halves of the Same Brain

If individual roles are converging, Foley sees no reason the same logic should stop at the boardroom door. His view on technical leadership is one of the more provocative in this report, and he offers it with a characteristic caveat.

"If you're a technologically orientated company, a CPO and a CTO don't seem like two roles to me. They seem like two halves of the same brain."

"If you're a technologically orientated company, a CPO and a CTO don't seem like two roles to me," he says. "They seem like two halves of the same brain."

He is careful not to overstate it. In regulated industries, roles such as the CIO exist for good reason, with clear accountability for data and infrastructure. Plenty of businesses will continue to run successfully with distinct product and technology leaders. His point is narrower, and harder to dismiss. "If I was writing down an org chart, I'd ask myself, why am I splitting those roles? And I'd need a really good answer. If there is one, great."

His reasoning draws on history. Early in his career, he worked alongside teams who called themselves rackers and stackers: IT specialists who bought servers, wired up switches and routers, and handed engineers an IP address and a set of credentials. Automation and then the cloud absorbed that work. The people did not disappear; their roles expanded and changed shape.

The same pattern has played out repeatedly since. Separate iOS and Android developers gave way to engineers expected to work across both. Front end, back end, cloud and QA specialists gave way to engineers who own a product from top to bottom. "I think in terms of outcomes," he says. "You want somebody to own the outcome. And I think it's the same for leadership. I don't know how you'd be just a technology leader anymore. That's a box I'd be nervous to be in."

He also recognises the lens he is speaking through. "We all see the world through our own biases," he says. "I see myself as commercially orientated, product orientated, with a technical and builder's background. So in my mind, I'm probably thinking, well, isn't everybody like me?" Then, with a laugh: "I'm known for two things. Having an opinion and being wrong. So take everything I said with a handful of salt. Not even a pinch. A big massive handful."

The honesty is disarming, but the underlying question deserves attention from every leadership team. As AI makes building, research and access to information dramatically easier, the case for purely horizontal roles weakens, at every level of the organisation.

More Swings at Bat

For organisations still early in their AI journey, Foley's advice starts with an admission. When he first saw the early GPT models, shortly before ChatGPT took off, he dismissed them as a nice demo. He now counsels balance in both directions.

"Don't dismiss it, but also do not take too deep a gulp of the Kool-Aid," he says. "It's a spectrum. If you find yourself on either end, you're probably wrong. There is value in the tool. There is value in the technology."

He adds a useful reminder that AI is a far broader field than the large language models currently dominating the conversation.

His central principle is to let the business problem lead. "Don't let the tail wag the dog," he says. "Start off with a problem that you need to solve, that you believe AI can help you with. Start with that hypothesis, define what success looks like, but also define what failure looks like. And just be experimental."

The length of an experiment, he argues, should be calibrated to the organisation. Two years is almost certainly too long. Two days is probably too short. The right window is the one that lets you learn quickly without betting too much on any single attempt. It is a philosophy he absorbed years ago from a product leader who believed nobody truly knows what will work until it touches a customer, and that every feature should therefore be treated as an experiment.

"AI is the same as any other technology," he says. "What is the business outcome you're going after, the problem you're trying to solve? Using AI to make that happen is an experiment. So build the experiment and don't be too precious about it."

Whether that happens through a dedicated innovation function or inside existing product teams is, in his view, a cultural choice each organisation must make for itself. What matters is cadence. Working with largely American colleagues, he reaches for a baseball analogy. "You want as many swings at bat as possible," he says. "If you run one experiment a week, you get fifty-two swings a year. If you run one a quarter, you get four. If you think of it as a probability game, the more swings you get, the more likely you are to hit the ball."

"You either win or you learn. Whoever keeps swinging will eventually hit the ball."

Failure, in that framing, is simply part of the process. Two of his five early projects have taken off, and he fully expects others along the way will not. "You either win or you learn," he says. "As long as you take the learnings to the next thing, you're improving your probability all the time. Whoever keeps swinging will eventually hit the ball."

AI Owns the Bits, Humans Own the Atoms

Asked whether we are heading towards utopia or dystopia, Foley gestures to the shelf behind him, home to a well-loved Blu-ray collection of Star Trek: The Next Generation. He is careful not to put dates on it, but he believes we have taken the first steps towards the world that show imagined. Not space travel. Abundance. His reasoning is grounded in economics. As populations grow and pressure for affordable food intensifies, agriculture will increasingly rely on autonomous systems, robotics and AI to remain viable. The energy demands of AI itself will push societies towards cheaper, more plentiful power. Advances in AI-driven protein modelling are already helping scientists screen drug candidates digitally before they reach the lab, raising the odds of success and, over time, bringing better treatments to market faster and more cheaply.

"Cheap energy plus cheap food equals what?" he asks. "If you keep stacking things on top, to me it's inevitable that you end up with abundance. Is that next year? No. The next decade? No. Is it in the next century? Definitely, in my opinion. My son is six months old. I think he might see it."

That optimism comes with a genuinely thoughtful question attached. If work no longer dominates our lives in the way it has for previous generations, where will people find their identity? It brings the conversation full circle, back to the identity questions facing designers, engineers and leaders today.

Foley's answer is captured in a line he has started using with friends. "On a long enough timeline, AI owns all the bits and humans own all the atoms," he says. The physical, tactile and imperfect will matter more, not less. Craft may return not as a living, but as a source of meaning. "You may become a woodworker, not because you need to be a carpenter to get a job, but because you like to build things."

He has acted on this himself. Over the past eighteen months, as his use of AI has deepened, he has moved deliberately back towards the physical: CDs, books and films on disc in place of streaming. "It felt sterile. It felt too perfect," he says. "Sometimes you need rough edges. You need a unique path. If every path is perfect, it's going to get boring."

He applies a similar discipline to how he thinks. His own most valuable use of AI is as a sparring partner rather than an oracle, something he talks to on evening walks to stress-test ideas and expose blind spots. "I'm not asking it for the truth," he says. "I'm asking it, where are my blind spots? What am I missing?" But he has also caught himself drifting towards whatever a model suggested first, and he is conscious of what that could mean for a generation growing up with an AI button always within reach.

"If you offload too much of it to the AI, you go limp. You're brought along by the tide, and AI is the tide. At some stage you need to be able to swim, so you can forge your own path."

"If you offload too much of it to the AI, you go limp. You're brought along by the tide, and AI is the tide," he says. "At some stage you need to be able to swim, so you can forge your own path. The balance is between, it's helping me, but I'm in control, versus I've gone limp and the thing is carrying me forward."

It is a fitting place to end. Foley is, by any measure, one of the more enthusiastic voices in this report. He has watched agents compress his own work by factors of two and three, helped turn thirty-day experiments into live products inside a global firm, and sees a future of extraordinary possibility. Yet his clearest message is not about the technology at all. It is about the people using it: commercially minded, experimental, willing to dream a little bigger, and determined to stay in the driver's seat.

Clive Foley is a Director of Engineering in the Innovation Centre at Marsh, where he leads teams turning AI- enabled ideas into validated products for businesses across the group. This piece is part of Signal 2026, Gathered & Found's series of candid conversations with the technology leaders shaping how organisations adopt AI.

‍

Let's stay connected

Join our newsletter and stay ahead!

X
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.

Have an important challenge that can't wait?

Tell us what you’re trying to achieve and what’s getting in the way.

We’ll help you work out the expertise and approach needed to move it forward
‍

Tell us about your challenge