Earlier this year, a Board Chair we know well spent a day at a conference on AI for Board Directors. There were about a hundred delegates, many of them with long careers behind them in technology, finance and regulated industries. Early on, the speaker asked the room to rate their own fluency with AI on a scale of 0 to 5. Zero meant never having used it, and 5 meant an experienced user working at the frontier of what the tools can do. Hands went up at 1 (uses for casual questions), and more went up at 2 (prompts in a structured way). Not one delegate placed themselves above Level 2.
It was only a show of hands, and self-assessment in front of peers can err in either direction. But these were Directors who had given up a day to the subject, and the ceiling (at Level 2 of 5) was a surprise. The issue is that the decisions now landing on Board agendas require an understanding above that ceiling.
Level 2 describes the deliberate prompter: someone who gives the model context, sets a format, iterates and gets good answers back. It feels like fluency, for those unaware that much, much more is in reach for all users. For many CEOs and Boards it is where AI literacy is assumed to stop. Yet Boards are now being asked to approve plans to automate processes across teams, give systems access to customer data and let software act without asking first. All of that is at Level 3 and beyond. A Board Director who has never been there is approving an architecture he or she has no first-hand basis to judge.
Part of the reason so many sit at Level 2 is that the frontier has moved rapidly. A year ago it took real skill to get a good answer out of a language model. Today the tools have absorbed most of that skill themselves. Claude, for example, can take a half-formed question, work out what is actually being asked and get somewhere useful without being told how to think. Competent prompting is now available to anyone with a phone or laptop and the ability to type or speak into it, so it no longer marks anyone out.
The conference’s show of hands proved the gap in AI vocabulary that we suspected sat at Board level and it raised an obvious next question: how fluent do Directors really need to be? And, standing in front of the mirror ourselves, how fluent do all the Perigon team need to be given our business involves working with Board and Executive teams and increasingly itself is built on AI foundations. So we borrowed the concept from the conference and designed our own AI Adoption Scale, from which to build training for our own team. You may find it useful too and you can access it here.
Levels 0 to 2 sit inside the chat window. Level 1 is a single prompt with the first answer accepted. Level 2 brings structured prompts, persistent context, uploaded documents and iteration, and it is the Level most self-described AI-savvy people mistake for the ceiling. Level 3 is the threshold: AI connected to a real system for the first time, an automation that runs without anyone starting it, a tool other people use. Level 4 is designing agents that work across several steps and tools, handle their own failures and know when to hand a decision back to a person. Level 5 is redesigning how the organisation works with AI embedded, including designing the frontier controls and governance and, critically, where AI should not replace humans versus where it should.
The cost of the senior team stopping at Level 2 is similar to asking them to approve investments and plans presented to them in a different language. A CEO or Board who has never built even a simple agent has no real basis for judging whether the technology team's AI roadmap is ambitious. They can't ask the important questions: which parts of a process is AI stepping into, connected how, checked by whom, failing safely in what way?
So where should a top team aim? Our view is that most Directors and Executives need to reach Level 3, and ideally Level 4. Currently, that means being able to build things using code or agents rather than simply have a conversation. That’s what we use at Perigon as the current benchmark for fluency in AI. We don’t think Level 5 (the ability to redesign entire systems to be AI native) is required of everyone, but the business needs at least one person working there, close enough to the frontier to see where it is heading.
In the same way that nobody became fluent in a language by reading a paper about it and signing off a budget for someone else to learn it on their behalf, AI fluency requires the individual to get their hands dirty. Explore it, learn how to push it to new levels, practise. All this really requires is time, freedom and access to a large language model of choice. Perhaps just ask the AI to design you a curriculum to get to the next level.
The goalposts will keep moving as the technology continues to evolve at breakneck speed. What counts as Level 4 today will look basic within anything from a couple of years to a few months. But a top team that holds itself at Level 3 or 4 as the scale moves should be well placed to govern AI and challenge the plans put in front of it. The collective profile is the best way to reassure that the necessary level of governance and oversight is in place for AI.
This whole article has been about AI but in fact none of this is really a question about AI. It is the question every new technology eventually asks of a Board: do you understand it well enough to direct it, or only well enough to be impressed by it? If your Board was asked for a show of hands on an AI fluency scale tomorrow, at which number would you reach a collective ceiling?
