Cognitive Crush
The new AI bottleneck - and the FOCCCUS Formula
Hello!
Apologies … It’s a been a while. I published The SPEED Book in January, and I promised to tell you more about it, and then … I had a wee hospital visit and some recovery. Sorry about that - I’m all good now.
I will share more about The SPEED Book, and the other book I wrote while I was recovering(!), very soon. Promise.
But today I want to share a very important “Bottleneck Guy Insight” that will help you look very clever … if your business is currently investing in AI coding tools, like Claude Code.
Because … guess what?
AI is going to really mess things up, if you’re NOT bottleneck savvy.
It’s taken me ages to find the words to say this, so if you have a few minutes it’d help me loads to know if the words resonate or not.
Cognitive Crush - a mini-essay
→ AI just made your developers twice as fast. Congratulations - you’ve created a brand new bottleneck.
Here’s what’s happening in company after company right now. Engineering teams adopt AI coding tools - Copilot, Claude Code, Cursor, pick your flavour - and developers speed up. Significantly. That part works.
But faster code means more decisions needed, faster. More features to prioritise. More designs to review. More trade-offs to evaluate. More ideas to say no to. All of that lands on your product managers, designers, and leaders.
The obvious fix? Give your product people AI tools too. Speed them up the same way you sped up engineering.
It helps. But it almost makes things worse. Because AI doesn’t just speed up their work - it multiplies the volume of options, analysis, and insights their brains have to juggle. Work that used to be spread across weeks is now compressed into days, if not hours.
I’ve decided to call this “Cognitive Crush” - the cumulative weight of all that AI-accelerated thinking, pressing down on the humans who have to absorb it. It’s not one big thing. It’s a thousand useful things, each one adding weight, until something gives.
Our dog trainer taught us something useful: if you want to exhaust the dog, don’t run them around the park - make them play brain games instead. Hide treats, make them solve problems. They’ll sleep for hours.
It’s the same with people.
Hard thinking is massively more exhausting than other work. And AI has just replaced a huge chunk of your product team’s work with pure, concentrated decision-making. Hour after hour. Day after day.
It’s like going from driving on quiet country roads to navigating a busy city. Same distance. Same car. But the cognitive load per kilometre is through the roof. You arrive exhausted, and you’re not sure why.
You can’t solve a brain bottleneck by throwing more AI at it. More AI just feeds it. Makes it worse.
The leaders who’ll navigate this well will start the same way you solve any bottleneck problem.
Step one: find it. And right now, for a lot of companies, it’s not where they think it is. It’s not in the codebase. It’s inside the heads of the people who have to think, decide, and prioritise.
Step two: manage it carefully. Which in this case means managing cognitive energy like the precious, finite resource it is.
Find then manage - this is my FOCCCUS formula, right?
The short version is: don’t make people’s brains explode.
Hope that helps.
That’s going up on LinkedIn this evening (but with fewer incomplete sentences, because LinkedIn seems to like those.)
Hope you’re doing well. Why not hit reply and tell me what’s up?
I really do like that :)
Clarke
Most important thing: you are back....
For AI, there will be a lot of new bottles uhh bottlenecks.
The is the joke that AI creating those could help to solve ....
he Cognitive Crush piece is doing something real, but there's a tension in it worth examining.
You're describing a bottleneck problem and reaching for bottleneck language — find it, manage it, FOCCCUS. That framing works for throughput constraints. A bottleneck in a production system is a queue that's too long; you identify it, you widen it or you slow the upstream. But cognitive load doesn't behave like a queue. You can't widen the human decision-making capacity the way you can add a server. And the decisions themselves are often not separable — the product manager can't just process prioritization tickets faster, because the quality of each decision depends on the cognitive state carried in from the previous one. Fatigue isn't just slowness, it's degraded judgment, which means the output of the bottleneck gets worse as the queue grows, not just slower.
The dog trainer analogy is pointing at something true — mental work is exhausting in a way that's invisible until it isn't — but it might be pointing at the wrong solution. You exhaust the dog with brain games so it sleeps. The implication is: give people less cognitive work, or space it out. But in the scenario you're describing, the volume of decisions is structurally determined by how fast engineering is shipping. You can't slow the dog down by making it nap. The features keep coming.
The more interesting version of the problem might be: AI has changed what kind of thinking is required, not just how much. Before, product managers were doing a mix of coordination, documentation, stakeholder management, and actual hard thinking. AI is automating the coordination and documentation faster than it's automating the hard thinking. So the ratio shifts — more of the remaining work is the expensive kind. That's not a bottleneck problem, it's a job composition problem. The role is being hollowed out from the easy end, leaving the hard end exposed.
Which suggests the intervention isn't managing cognitive energy as a finite resource to be rationed — it's redesigning what the role actually does when the easy parts are gone. What does a product manager do when AI handles the documentation, the analysis synthesis, the stakeholder update drafts? Either the role becomes pure judgment and taste, which is exhausting in the way you describe, or it becomes something else entirely that we don't have a name for yet.
Where does FOCCCUS point when the bottleneck isn't a queue but a job description?