Last week, we gathered around a table at Claridge's with senior technology and business leaders for our second breakfast in partnership with Opulent Mind.
The room was made up of mixture of CTOs, COOs, founders, product leaders and investors. There was a real balance of Startups raising their first round sitting next to people running listed businesses.
When we ran the previous session, back in March, we said everything was moving so fast that we should do another one before the summer. Four months later, we did. And the conversation had moved on, noticeably.
The session ran under Chatham House rules, so nothing here is attributed to anyone. But the themes were as follows:
I opened with a stat from Deloitte's State of AI in the Enterprise report: 66% of organisations report efficiency gains from AI, but only 34% say they're actually reinventing products or processes.
The early-stage founders in the room aren't retrofitting AI into anything, they're designing operating models from scratch that assume it. Different roles, fewer hires, products conceived differently. As one put it: it doesn't make sense to design a traditional operating model anymore.
The sharpest example came from financial services operations. The efficiency mindset says: automate the back-office portal. The reinvention mindset asks: why do you have a portal at all? Why not an MCP server connected to your data lake, with agents reacting to real-time information?
Same problem. Completely different answer.
And one comment stuck with me: a CEO cares more about making more money than about continuing to squeeze the lemon on cost. The organisations thinking about AI as a way to fundamentally change what they sell - not just how cheaply they make it - are the ones that will thrive.
In March, the CTOs around the table were spending most of their time managing board expectations. That hasn't gone away, but the conversation has matured into something more specific: the economics of people.
You might reduce headcount but you won't reduce cost pro rata. The pattern several leaders described was that you don't need 50 engineers on £45k anymore. You need 15 on £120k. Content creation becomes content curation, and curation demands more skilled people, not fewer.
There's a quality trap hiding in there too. Experienced people using AI get exceptional results because they know what good looks like. Hand the same tools to a mid-level team and the quality bar drops as they can't judge what the AI is producing. The value of expertise hasn't fallen. It's been magnified.
The likely losers? The big offshore body shops. Several people described moving from large offshore teams to smaller, more senior onshore and nearshore ones, and once you add up tooling and higher salaries, the total cost isn't far off. But the capability is.
The board conversation generated the most knowing laughter of the morning.
Boards still want the gold-plated waterfall plan: three-year value, milestones, little black diamonds on a Gantt chart. And the world has simply stopped working that way.
What leaders described instead was, ironically, agile in the boardroom. Run ten proofs of concept. Expect five to fail. Pick two of the survivors. Educate the board that fail fast, fail often is a strategy, not an excuse.
The cultural contrast with the US came up more than once: mandated adoption and experimentation over there, hesitancy and ivory-tower governance here. One guest put it memorably: you don't want one oil tanker, you want ten speedboats.
And the culture point cut deepest: people will take risks, break things, and ship the 80% solution only if they know it's safe to do so. In too many organisations, failed experiments get quietly buried instead of openly reported. That's a leadership problem, not a technology one.
This is where the conversation got existential.
Previous transformations asked people to change how they work. This one asks knowledge workers to put what's in their heads, the very thing they've banked their careers on, onto the machine. Across professional services, law, finance, HR. Front office included.
That's a hard sell, and pretending otherwise doesn't help. The most effective approach anyone described was radical honesty with teams: we don't expect you to be here for life; for the next three years you'll do cutting-edge work that strengthens your CV. Treat people like adults and many will take that deal.
The M&A perspective was one of the most striking of the morning.
Deals that started before Christmas hit the buffers after it, including one pulled by the investment committee a week before signing. For tech-enabled businesses in the lower mid-market, the AI question isn't discounting valuations. It's driving 99% no-bids.
But here's the flip side: businesses with good economics and genuine AI durability are attracting inflated prices, because there are so few of them.
And the opportunity investors are hunting for? Revenue transformation. Take £5m of a services business's £15m revenue and turn it into productised technology revenue, and you've multiplied its enterprise value without making a penny more. Services revenue trades at 3–4x. Product revenue at 10–12x. Same money, very different business.
We closed by asking what the conversation will be in three to six months.
One guest had spent a day at their kids' school getting pupils to build and ship a game with AI and teachers loved it, and they don't have the tools yet. Another described a network of 300 schools running full AI learning loops: every pupil's exam history feeding personalised homework that adapts monthly. Extraordinary stuff.
But the open question hanging over all of it: we know what happens when you give an LLM to someone who knows what good looks like. What happens when you give it to someone who doesn't?
Two other threads to watch. First, the regulated-industry gap, surface-level applications are easy, but the deeper you go into the product, the harder the regulatory boundaries bite, and there's still a suspicion that some firms are hiding behind compliance rather than being blocked by it. Second, preserving critical thinking. If you're not mindful of the boundary between your knowledge and the AI's, you're outsourcing the very judgement that makes you valuable.
In March, the theme was honesty with boards. This time, there was a shift in the question itself, from "how much more efficient can AI make us?" to "what should this business even look like now?"
Efficiency is a given. It's table stakes. The leaders who will win are the ones redesigning what they sell, how they're valued, and who they hire and having uncomfortable, honest conversations with their boards and their people along the way.
We'll run the next one after summer. Judging by the last four months, it'll be a different conversation again. That's the whole point.
If any of this resonates, or you disagree with all of it, I'd love to hear from you. These conversations are better with more perspectives in the room.