Last week, we hosted a dozen senior technology and business leaders for a breakfast roundtable exploring their approaches to navigating the AI change in enterprise technology. CTOs, COOs, a chief digital officer, a chief data and AI officer, from across financial services. Insurance, banking, asset management, fintech.
I love running these events. I’ve done enough of them now to know that the best ones don’t follow a script – they follow the room. This one didn’t disappoint. What stood out most was how honest everyone was. Not the polished conference-stage version of AI strategy, but the real one: the tensions, the contradictions, the things that are working and the many things that aren’t yet.
The session ran under Chatham House rules, so I won’t be attributing anything to anyone. But the themes that emerged were as follows:
Everyone’s doing AI. Almost nobody’s reimagining anything.
This was the first thing that became clear. Every organisation in the room is running AI initiatives. In fact, most are running several. Efficiency gains are happening: processes that took an hour compressed to five minutes, manual triage replaced by intelligent routing, that sort of thing. The quick wins are real.
But when the conversation turned to genuine reimagination – rethinking propositions, restructuring operating models, building something that couldn’t have existed without AI – the room went quieter. A few people were honest enough to say: we’re not there yet, and we don’t have a structured way of getting there.
The contrast with the startup world is stark. AI-native companies are hitting hundreds of millions in revenue with teams you could fit around a single table. That’s not immediately relevant to a regulated financial services firm, but it does tell you something about what’s technically possible when you don’t have 20 years of legacy holding you back.
Several people made the point that regulation acts as a temporary moat – it buys time, because new entrants face the same compliance burden. But it’s a double-edged sword. The same governance structures that protect you also slow down your ability to respond when something genuinely disruptive does appear on the horizon.
The most interesting observation was about where the pressure is coming from. For many in the room, the AI conversation is being driven less by a clear commercial thesis and more by board anxiety. One person described it well: the board reads about the latest model release and wants to know why the company hasn’t deployed it yet. The CTO’s job is increasingly to channel that energy into something useful rather than just react to it.
We deliberately pushed the group past pilots and innovation labs. Where is AI actually in production, doing real work, with real accountability?
The answer: in a handful of places, and they all follow a similar pattern. You pick a single process where you have deep data. You build an agentic workflow that handles the heavy lifting – reading documents, extracting information, applying rules, calling pricing APIs. You keep a human in the loop as a quality gate. And you plan, eventually, to reduce human oversight as confidence grows.
The gains from this pattern are significant. People described processes that used to take the best part of an hour being reduced to minutes, with quality that holds up. But nobody in the room has yet taken the step of removing the human entirely. That’s partly a compliance question, but it’s also a trust question. The technology may be ready before the organisation is.
One point that landed well: measuring token usage is measuring an input, not an output. Companies that get excited about how much AI they’re consuming, rather than what it’s producing, are going to come unstuck. The value metrics haven’t changed – you either grow revenue or reduce cost. AI is the means, not the measure.
This was the topic that generated the most heat. And honestly, no one had a clean answer.
The pattern everyone recognises is this: senior people partnered with AI agents are where the real productivity gains live. The most experienced underwriter, engineer, or recruiter – working alongside an intelligent agent – is dramatically more effective than they were alone. Not because AI replaces their judgement, but because it handles the work they were worst at and frees them for the decisions that actually matter.
The uncomfortable corollary is the junior pipeline. If AI handles the repetitive work that graduates used to learn on, how do you train the next generation of experts? One person pointed out that hiring an assistant underwriter barely makes sense anymore when the agent can do that job. The same logic applies to junior actuaries, paralegals, and arguably junior developers.
Not everyone agreed. Some argued that the new generation will arrive already fluent in AI-native working – they won’t need the old apprenticeship model because they’ll learn faster in partnership with the tools. Others worried about the current mid-career cohort who trained the old way and are now caught between two worlds.
There was also a genuinely surprising insight about customer behaviour. The assumption that younger customers want full automation turns out to be wrong – or at least more complicated than anyone expected. In some sectors, it’s younger customers who want a human to hold their hand through a big financial decision, while the older generation is more comfortable self-serving. Worth remembering before you automate your entire front line.
We ended the morning on leadership – specifically, what the CTO or CIO role actually looks like from here.
The consensus was that technology is becoming a hygiene factor, much like Excel did a generation ago. Nobody asks whether a finance hire can use a spreadsheet. Soon nobody will ask whether a developer can work with AI. The differentiating skill for a technology leader is shifting away from technical depth and towards people leadership, commercial awareness, and the ability to educate and influence the board.
Several people made the point that their most important responsibility right now is not AI-driven transformation – it’s risk management. Keeping the firm solvent. Keeping data secure. Making sure the defences evolve as fast as the threats do. There was a candid admission that bad actors are adopting AI tools faster than most enterprises’ own security is keeping pace. That’s a sobering thought.
The capital allocation question came up repeatedly. Traditional five-year investment horizons simply don’t work when the technology landscape shifts every quarter. One leader described telling their board: I can’t promise this platform will still be the right answer in two years. I might come back and ask to rebuild it. That kind of honesty takes courage, but it’s what the moment demands.
There was also an interesting tension between the CIO and CTO roles. One view – which I found compelling – is that these roles are splitting more cleanly than they used to. The CIO becomes the operational and people leader; the CTO becomes the forward-looking, commercially-oriented executive who sets direction and influences the board on where to invest. Whether that split happens formally or informally, the cognitive load of trying to do both is becoming untenable.
If there was a single thread running through the morning, it was this: the most important work in AI right now isn’t happening in San Francisco. It’s happening in the messy middle – inside regulated enterprises where leaders are trying to move fast without breaking things that matter.
The people navigating this best are the ones who resist the hype, focus relentlessly on production, invest in their people, and stay honest with their boards about what they don’t yet know. That last part is harder than it sounds. Nobody wants to stand in front of a board and say “I’m not sure yet.” But the leaders who can do that – and follow it with a credible plan for finding out – are the ones I’d back.
We’re planning to run the next one of these in a few months. Had we held this breakfast three months ago, the conversation would have been quite different. It’ll be different again next time. That’s the whole point.
If any of this resonates – or if you disagree with all of it – I’d love to hear from you. These conversations are better with more perspectives in the room.
Red Badger is a digital product consultancy that works with enterprise organisations to build software that matters. We regularly host closed-door conversations for senior technology leaders. If you’d like to join a future session, get in touch.