Every AI tool you bring into your business this year will do one of two things. Make your people sharper. Or make them smaller.
Almost nobody buying these tools is asking which one they're getting.
Last month at Oxford's Global Leadership Forum, Philipp Koralus (the university's professor of philosophy and AI) opened his keynote with a show of hands. Who has agents running on their behalf right now? Who has leaned on AI for a real judgement call at work? By the second question nearly every hand in the room was up, including mine. A room full of senior leaders, every one of them already outsourcing judgement, none with a framework for where it ends.
The two failure modes
Behind every AI tool is a decision about what it does with your judgement. Almost all of them pick one of two wrong answers.
The first kind stays out of your judgement entirely. This is most of the AI bolted into workplace software today. It summarises the meeting, drafts the reply, tidies the thread, and leaves every decision exactly where it found it. Useful, harmless, forgettable. Usage spikes at rollout, fades within a quarter, and nobody fights for the licence at renewal.
The second kind takes your judgement over. This is the agent wave arriving now: the calendar that decides what you work on next, the inbox agent that sends the follow-up instead of drafting it, the meeting bot that turns a half-formed thought into an assigned task. Quietly, and on your behalf. People love these tools, and they don't stop loving them. That's the problem. The skill the tool replaced fades, until doing without it no longer feels like an option.
You already know how this goes, because it's already happened with GPS. A decade of turn-by-turn navigation, and most of us can no longer find our way across our own city without it. Fine for driving. Now imagine the same thing happening to prioritisation, follow-through, and decision-making.
Look at the AI already inside your company, then at the agents in next quarter's vendor pitches. Almost all of it is one or the other.
Why the second kind is worse than it looks
If, like me, you're old enough to have owned a paper map, losing your sense of direction to GPS is losing a skill you already had. Anyone younger never built the skill at all. That's the deeper cost Koralus pointed to: handing over your judgement stops you building the judgement you don't have yet.
He demonstrated it with a well-known experiment. Give half the people in a room a coffee mug, then let everyone trade. The mug owners immediately demand twice what buyers will pay… a bias called the endowment effect. But keep the market running, round after round, and the bias disappears. People learn by doing. They get better at valuing things by practising valuing things.
Bad judgement isn't a fixed trait. It's trained away by practice.
A system that makes the choice for you removes exactly that practice. The work gets done. The learning stops.
Now scale that across your organisation. Picture every manager whose AI drafts the hard message, makes the prioritisation call, decides what matters this week. The dashboards will show time saved. What they won't show is managers who've stopped getting better at the one thing you pay them for.
And if you run the business yourself, it's starker: the judgement being handed away is yours, and everything downstream of you depends on it.
The design constraint I took home
I'm building readywhen, a senior AI employee for owner-operators. The week after the keynote, one line went into our strategy document: help people move forward without taking over. Make people more capable, not more dependent.
That's not a mission statement. It's a design constraint, and it bites. It means the product has to earn the right to be opinionated. Show why something matters before acting on it. Keep your own words visible instead of paraphrasing them into something smoother. Ask before turning an uncertain thought into a commitment.
The kind of line I want readywhen to say to me: "You have 14 open follow-ups from last week. Three are customer-facing and two are blocking Sam on your team. I'd clear those before polishing the internal deck." That's the whole job. Clear what doesn't need me. Name what does. Get out of the way. What it doesn't do is take the deck away from me, or pretend it knows my customers better than I do.
The version that keeps people sharp is slower to build and harder to sell. None of it reads as magic in a product demo. All of it is the difference between a tool people trust with their working life and a tool they stop opening.
The question to ask before you buy
Koralus offered a test for any decision-support tool, and it's the most useful sentence I took from the whole event: how will this interact with my people's ability to adapt and learn?
Not "how much time does it save." Not "what's the adoption rate." Whether the people using it are getting sharper or smaller.
If you're rolling out AI across your team this year, put that question to the vendor. Watch how they answer. Most can't, because they've never thought about it. The ones who can are the ones building tools your best people will still be using, and still be better for, in five years. And if there's no vendor and no procurement, just you and a credit card, put the question to yourself before you switch the thing on.
What you think people are for
Whether any of this matters comes down to one belief.
If people are just labour, it doesn't. Automate the judgement, take the savings. Some companies will run that play, and for a few quarters the dashboards will applaud.
I'm betting the other way. AI is making the routine parts of every business cheap and identical. What it can't commoditise is the handful of calls that actually decide the outcome: what to build, which customer to believe, who to hire, when to walk away.
Those calls are the game now. A team that keeps making them gets sharper every quarter. A team that hands them to software doesn't notice what it's losing until a call arrives that the software can't make.
readywhen is that bet in product form. Clear the work that doesn't need people, so they get more practice at the work that does.
Koralus gave me the language. The belief was already mine: people are capable of far more than the work in front of them usually asks for. If that's wrong, buy whatever saves the most time. If it's right, the winners of the next decade will be the companies whose people kept getting sharper while everyone else's got smaller.
Which bet is your company making?