A year ago, the interesting question inside most companies I worked with was how to get more out of the model — longer context, cleverer prompts, more tokens pushed through a bigger pipe. Tokenmaxing. It felt like the frontier. It isn’t anymore.
AI adoption didn’t creep in gradually. It went from a promising experiment to a company-wide expectation in barely any time at all, and it stopped being an engineering initiative the moment finance, legal and operations all started asking for it by name. That changes the problem. When the tools were scarce, access was the advantage. Now that everyone has them, access is table stakes — and the thing that’s actually scarce is agreement on what you’re using them for.
The gap I keep walking into
It looks the same almost everywhere. Teams want the room to move fast. Leadership wants to give it to them. And sitting in the space between the two is a question nobody has quite answered out loud: what business outcome are we actually aiming for? Not which tools get rolled out, not how many seats get licensed — the outcome.
It’s never a shortage of effort or ambition. It’s that agreement hasn’t caught up with the pace of change. And without it, spend doesn’t reliably turn into results. You get motion that looks like progress and a P&L that doesn’t move.
The gap most companies have isn’t access to the tools. It’s turning that access into changed behaviour and outcomes you can measure.
Why it’s a judgement problem
Here’s the uncomfortable part: closing that gap is mostly a judgement challenge, and judgement is exactly the thing you have no time for when you’re deep in the day-to-day of getting work done. There’s rarely a moment to step back, sit in the ambiguity, and work out what the new playbook should actually be. So the old one keeps running, a little faster each quarter — the hamster wheel.
The work I care about is getting a business off that wheel and onto a flywheel: one that keeps accelerating because the team is doing the higher-value work, and the tools are working for them, not the other way round. That switch is never a technology decision. It’s a decision about what to stop doing, what to commit to, and who owns the outcome.
Where AI earns its place
I’m not anti-tool — far from it. AI shows up wherever it earns its place in this work: sharpening discovery, taking busywork off the team, buying back the time for the judgement calls that actually matter. But it earns that place against an agreed objective, not for its own sake. When the objective is clear, the question of which tool, how much context, how many tokens mostly answers itself.
So tokenmaxing is dead, and good riddance. The frontier moved. It’s no longer about squeezing more out of the model — it’s about the far less glamorous, far more valuable work of getting a business to agree what it’s for, and then doing the work to get there. That’s the gap I spend my time closing.
