Everybody thinks they are behind

Almost every large company believes it sits in the bottom quartile of AI adoption for its industry. The arithmetic cannot work. But the feeling is near universal among all the companies we engage with, and the feeling surfaced again within the first half hour with the senior leaders we hosted for dinner at Imperial College London to talk about the enterprise AI capability gap.

The group was mixed: banking, luxury retail, mass retail, travel, entertainment, even pet food, alongside Imperial academics including Professor Chris Tucci, Vice Dean of Education.

Chatham House rules applied, so what follows is unattributed. It is also unvarnished, which was rather the point of the evening.   

 We had some really candid views on the night. One leader prefaced their comments with "this could get me fired", then described what happens when a group CEO simply tells everyone “to do AI”. It’s the classic licence requests with no guardrails, enormous energy and intent with no direction, and results in their teams "building incredible agents to manage spreadsheets, rather than removing the spreadsheet". This scenario was greeted with many nodding heads.

A downstream symptom of that mess also came up - anyone trying to slow things down and ask what the process should be is too often labelled as a blocker. 

A phrase from earlier that day kept resurfacing: ‘cognitive surrender’. By outsourcing our thinking to the agents and the LLMs, are we giving up the most important capability we as humans need to direct the machines - our critical thinking? One organisation had responded by teaching critical thinking before it rolled out a single AI tool, as they recognise the human skill needed now is no longer producing an answer but rather judging one.

The apprehension around cognitive surrender appeared throughout the evening… Generative AI generates, but organisations also need evaluation, integration and collaboration, which is why so many leaders report individual productivity gains that refuse to show up in organisational output. The backlog of work gets bigger. The words multiply. The bottleneck shifts. Decisions are not improving and the impact is not showing up in the numbers. We had the benefit of the academic view - and they gave the frustration a longer history.

Lisanne Bainbridge described the irony of automation back in 1983: automate the easy 80% of a task and humans get worse at the hard 20% that remains. A sharper worry emerged too. If every firm reasons with models trained on broadly the same data, their thinking converges. For one investment leader, that is not a productivity gain at all - it is correlated thinking, and correlated thinking is fragile. One guest offered the most quotable version: you can hike to the top of a mountain or take a helicopter, and you’ll arrive at the same place either way, but you do not arrive as the same person.

And then there is the question nobody around any table has cracked: juniors. Senior people know when not to use AI because their decades of slow, unautomated work taught them what good looks like. Junior employees are being asked to develop that judgement while the work that used to teach it is automated away.

It would be easy to read all this as gloom. Mostly, it’s realism. Leaders and organisations are shaping what AI means for them and learning as they go.

There are wins, there are challenges. What might be missing is the learning feedback loops that make everything better as we progress. 

What struck us most was how much practical, tested experience was already in the room. One adviser described a CEO who runs a show-and-tell at four o'clock every Friday, demonstrating his own AI experiments, failures included, because nothing gives an organisation permission to learn like watching the most senior person get it wrong in public. Another chief executive reportedly asked everyone to write "use AI" on a yellow sticky note and keep it on their laptop for a month. An adoption tactic that is crude, cheap, and effective.

There were harder-edged lessons too. One consultancy described abandoning the classic business case in favour of staged, venture-style funding: seed money for a proof of value, another round for a working concept, with permission to stop at any gate and no shame attached. VC expects a failure rate. The same firm now budgets AI transformation at roughly 30 per cent technology and 70 per cent people, an inversion of the digital-era ratio that drew agreement around the table. And the proof points are arriving. One process-embedded sales agent, which enriches and prioritises leads and then briefs the salesperson with a short generated audio summary on the drive over, is delivering a five to six per cent sales uplift in European trials.

The evening ended where the best conversations do, with the disagreement rising rather than settling. The final exchanges were not about tokens or use cases but about responsibility: what AI means for entry-level jobs, the wastefulness and environmental impact as we accelerate adoption and whether leaders are asking what all this capability is really for. One guest spoke about her ten-year-old daughter worrying about what work will be available to her. Another challenged the room's assumptions about data centre water and energy use with relish.

Two years ago, a dinner like this would have been all theatre: whispered anecdotes, borrowed conviction, everyone performatively demonstrating their 10x strategy. This conversation felt different and useful. A group of leaders running some of the most complex organisations in the country compared notes on what is working for them, what is failing, and what they genuinely do not know. We hope they left with perspectives they can use to reframe and ignite the work to be done. 

Everybody thinks they're behind. On the evidence of this evening, the ones taking time to explore, to indulge in thoughtful leadership and really grapple with uncomfortable questions are further ahead than they know.