AI is useful. It can help a team think, write, analyse, and synthesise at a speed that would have seemed extraordinary a few years ago. It can also hand you a set of OKRs that sound finished before anyone has had to advocate for their own beliefs and be curious about the beliefs of others.
That second trick is the dangerous one.
In AI wabi-sabi I wrote about what gets lost when we sand off the imperfections. One of those imperfections is how bad we are at uncomfortable conversations. We avoid challenging an assumption, questioning a priority, or admitting we might be wrong. Polite agreement is easier than a useful argument.
Those arguments are where organisations learn.
Kim Scott’s Radical Candor is a reminder that caring about someone and challenging them belong in the same conversation. Peter Senge, in The Fifth Discipline, makes a related distinction. In a debate, you defend a position, and try to win. A discussion is more inclusive. we we advocate and share multiple views and ideas. Dialogue is different again—slower and richer. Participants in a dialogue balance advocacy with curiosity, suspending assumption long enough to listen and really investigate what you and the others believe and why. If you’re lucky, you leave with ideas, solutions, and genuine understanding that no individual walked in with.
Nick Cave put the creative version of this plainly. A song written in his style can sound right and still be empty, because the struggle that would have produced it never happened.
Organisations are doing the same thing with OKRs.
A model can draft beautifully worded objectives, generate a dashboard, summarise the weekly update, and write a narrative about the quarter that would pass in a steering group. None of that tells you whether anyone in the room believes it. Senge describes levels of commitment, from non-compliance and grudging compliance, through formal compliance and enrolment, up to genuine commitment. There is a long distance between a team doing something because they were told to, and a team doing it because they decided it mattered.
Formal compliance is what a polished OKR often is. The words are clean. The key results have numbers. The slide looks decided. Then, in the check-in, nobody will say the objective is the wrong bet or that the key results don’t actually offer the proof a team needs. Confidence stays high because lowering it would start the conversation everyone has been avoiding. Sharing that number honestly only works if someone is willing to say it out loud.
At their best, OKRs are not a framework for setting targets and measuring progress. They put two things in the same room: the collective intelligence to find where the real leverage is, and the collective will to do something about it. That takes a lot of conversation. What matters. What is actually possible. What we are prepared to commit to. What we are getting wrong. What has to change before the next cycle.
AI can help you prepare for those conversations. It can lay out the evidence, surface a dependency you had missed, and offer a phrasing when the room is stuck on the words. Use it for that. Setting the OKRs is still a human argument about what you will and will not pursue.
It should not have the argument for you.
As producing the information gets cheaper, the scarce thing is a conversation that is honest, challenging, and able to produce a decision nobody brought into the room. The question isn’t how we use AI to make our OKRs better.
It’s how we use AI to make the conversations around our OKRs better.
Because that’s where the real intelligence is. And that’s where the real commitment comes from.