I was in a conference room at GHD’s Perth offices in April 2016, answering a question from the floor. I’d been introduced a few minutes earlier, slightly too generously, as “the secret weapon future proofing business,” which is the kind of line you learn to accept gracefully and never quite believe. The room was full of people from industry associations and universities, the usual mix at those events, and someone near the back wanted to know how organisations were meant to keep up with the pace of technological change without being run over by it.

I remember the answer better than I remember the question. I said the mistake people were making was thinking the problem was the technology. It wasn’t. The problem was outputs, not inputs. We had spent a decade getting extraordinarily good at putting new capability into organisations: new tools, new platforms, new systems, new automation. What we hadn’t built, anywhere near as well, was the human capacity to decide what any of it was actually for. I gave that capacity a name that day, because it needed one and didn’t have one yet. I called it Artisanal Wisdom: the judgment, the questioning, the intuition that turns raw capability into something worth having. I said it wasn’t about a loss of human control over technology. It was about whether we’d bother building the part of ourselves that knows when to use it and when to turn it off.

Ten years later, I want to tell you what’s arrived to confirm it. Not a trend. Not a prediction. A letter.

The letter

On 28 July this year, 1,178 employees from OpenAI, Anthropic, Google DeepMind and Meta AI signed an open letter asking the US government to support the development of something they called a pacing mechanism. Read the letter and you notice quickly what it isn’t. It isn’t a call to stop building. It isn’t an activist petition from outside the industry, wondering aloud whether AI has gone too far. It’s the opposite of that. It’s written by the people who are, right now, building the most capable AI systems that have ever existed, and its request is precise: build the technical and governance infrastructure that would let someone, somewhere, coordinate a verifiable slowdown if the pace of AI development ever outstrips the pace at which humans can safely oversee it.

Read that sentence again. These are the researchers and engineers who understand exactly what they’re building, better than anyone else on the planet understands it, and they’re telling their own governments that the thing missing isn’t more capability. It’s the capacity to judge, in time, what to do with the capability already in hand.

That’s not a new problem. I gave it a name in a function room in Perth a decade ago, because I’d already watched it happening for years by then, and I could see it wasn’t going away. What’s changed isn’t the problem. What’s changed is who’s naming it now, and how publicly they’re prepared to do it.

Two maps of the same territory

Most leadership teams I talk to are still reading the pacing letter, if they’ve read it at all, as an AI adoption story. Something about safety, something about regulation, filed mentally next to every other AI governance headline of the last two years. That’s an understandable read. It’s also the wrong map.

The people who wrote the letter aren’t describing an adoption problem. They’re describing a coordination problem. The distinction matters more than it sounds like it should, because the two maps point to completely different sets of decisions.

If you think you have an adoption problem, you invest in rollout: training programs, change management, tool selection, usage metrics. All useful. None of it touches the actual gap.

If you understand you have a coordination problem, the question changes. It’s no longer “how do we get more people using this well.” It’s “who, in this organisation, has the standing and the judgment to decide what this capability should and shouldn’t be allowed to do, and how fast can that judgment actually move.” That’s a different investment entirely. It’s not a training budget. It’s an authority map.

I built a framework for exactly this question after the Who Decides research in 2025, because boards kept discovering, usually mid-crisis, that nobody had actually decided who was allowed to decide. I called it Decision Trust Zones: a way of mapping, before you need it, where in an organisation a given class of AI decision should sit, and on what basis. The pacing letter is the frontier-lab version of the same discovery. Different scale, same architecture. Capability arrived faster than the authority to govern it did, and everyone downstream is now trying to retrofit judgment onto a system that was built to optimise for output.

What “outputs, not inputs” actually meant

I want to go back to 2016 for a moment, because the phrase deserves more than a passing mention, and because I think it’s more precisely correct now than it was then.

“Inputs” is everything you can measure going into a system: capital, compute, data, headcount, hours. “Outputs,” in the sense I meant it that day, wasn’t the product coming out the other end. It was the human judgment applied along the way: the questioning, the pushback, the moment someone in the room says “we can do this, but should we, and what happens three steps downstream if we do.” Organisations had gotten extremely efficient at measuring and increasing inputs. Almost nobody had built comparable infrastructure for the output that actually determines whether the input was worth anything: the quality of the thinking applied to it.

I don’t think this is a philosophical distinction. I think it’s a diagnostic one. If your organisation can tell you exactly how much you’ve spent on AI tools this year and can’t tell you who is accountable for the judgment calls those tools are now making inside your business, you have an inputs organisation. Data is not the same thing as knowledge. Knowledge is not the same thing as judgment. And judgment, not capability, is the layer that was always going to be the constraint. It just took until frontier researchers were writing letters to their own governments for that to become impossible to ignore.

There’s a hierarchy hiding inside that sentence, and it’s worth sitting with for a moment because most organisational thinking skips straight past it. Data is the raw material. Knowledge is data that’s been organised into something you can act on. Neither of those is what’s actually scarce right now. What’s scarce is the layer above both: the judgment that decides which knowledge matters, in which situation, and what to do about it before the moment has passed. Organisations have spent a fortune getting extremely good at the first two layers and have barely noticed the third one exists as a thing that needs deliberate investment at all. You can buy data. You can license knowledge. You cannot purchase judgment off a shelf, and I’d argue most boards haven’t yet worked out that this is the actual shopping list.

The ripple, three moves out

The first-order consequence of the pacing letter is obvious: it’s a governance story, and it will be covered as one. Regulators will respond, or they won’t. Frameworks will be proposed. Australia has already placed an Office of AI inside the Department of the Prime Minister and Cabinet, coordinating policy across investment, data, copyright and workforce, which is a first-order response to exactly this kind of signal.

The second-order consequence is closer to home, and it’s the one most boards haven’t looked at yet. If the people at the absolute frontier of capability are asking for coordination infrastructure they don’t currently have, the organisations several rungs down the capability ladder almost certainly don’t have it either, and haven’t been asked the question in a way that made the gap visible. Most leadership teams have spent the last two years building input infrastructure. Very few have spent equivalent effort building the judgment infrastructure that decides what the input is for.

The third-order consequence is the one that actually worries me, and it’s about people, not systems. When an organisation has strong input capability and weak output judgment, the gap doesn’t stay empty. It gets filled, quietly, by default: by whichever function happens to hold the budget, by whatever the vendor ships as a standard setting, by whatever decision got made eighteen months ago and nobody has revisited since. None of that is malicious. It’s just what happens when pace exceeds the time available for anyone to deliberately decide. The HUMAND framework exists because I kept watching this exact failure mode repeat across sectors: work getting allocated to human, machine or AI not because anyone chose deliberately, but because nobody was assigned the job of choosing.

The Who Decides research made this visible at scale. Across finance, healthcare, education and government, the sectors drew their lines on AI authority in genuinely different places, for genuinely defensible reasons. What far fewer of them could show was a documented, deliberate answer for why the line sat where it did rather than somewhere else. A regulated sector under obvious scrutiny, healthcare say, tends to arrive at that answer properly, because the stakes force the conversation into the open early. A business where the AI use looks smaller and lower stakes is far more likely to have a policy nobody has opened since it was written, and a decision quietly being made by a tool rather than a person. The frontier researchers writing the pacing letter have the advantage of knowing exactly how consequential their work is, which forces the judgment question into the open. Most organisations don’t have that advantage. Their AI use looks smaller, feels lower stakes, and that’s exactly why the authority question gets skipped rather than answered.

What this actually asks of leaders

Foresight isn’t a forecast, and I’m not offering one here. What I’m pointing at is a discipline of judgement that has to be practised in real decisions, under real pressure, and it doesn’t get built by accident.

If you’re leading an organisation right now, the pacing letter isn’t really asking you a technology question. It’s asking whether you know, today, who in your organisation has the standing to say no to a capability your business has already paid for. Not eventually. Today. If you can’t answer that in one sentence, you have the same gap the letter is describing, at a smaller scale and with lower public stakes, for now.

The work isn’t complicated to describe, even if it’s slow to do properly. Map where AI-related decisions actually get made in your organisation, not where the org chart says they should be made. In almost every business I’ve worked with over the last two years, those two maps have turned out to be different documents, and the gap between them is exactly where the judgment problem lives. Find the places where a decision has defaulted to whoever holds the budget or whatever the vendor shipped, rather than being deliberately assigned. Then decide, on purpose, who should be making that call, and give them the standing and the time to make it properly. That’s not a training program. It’s an authority map, built before the moment you need it rather than during the moment you’re arguing about it.

None of this is a criticism of the pace itself. Pace isn’t the enemy. I’ve never believed slowing down is the answer to feeling overwhelmed by change, and I don’t think the 1,178 people who signed that letter believe it either, which is precisely why they didn’t ask for a stop. What they asked for was the capacity to coordinate, if coordination is ever needed, before it’s needed. That’s a different ask entirely, and it’s one most organisations could make of themselves right now, at a scale that’s actually manageable, instead of waiting until the scale is frontier-lab and the stakes are civilisational.

I said in 2016 that the question was never the technology. It was outputs, not inputs. Ten years on, the people building the most capable systems in history have confirmed it in the most public way available to them, in a letter to their own governments. The technology was never going to be the hard part. It never is. The hard part was always whether we’d build the judgment to go with it in time to matter.

Choose Forward.