Answerable for What You Didn't Make.

Last week I let an AI take the first pass at something I’d normally grind through myself. It came back fine. Useful, even. And I sat there with twenty unexpected minutes and didn’t know what to do with them.

The machine doesn’t do the work badly. It does the work, hands it back, and leaves you holding time you forgot how to spend.

So I filled it. There’s a junior designer on my team who’d been stuck, and instead of jumping in to fix it, I pinged her on Slack and asked if she had ten minutes. We got on a video call and I asked questions until she could see the shape of the thing herself. That used to be the job. Somewhere in seventeen years it became the thing I did if there was time. I’d been able to spend the time on the person, instead of a deck — and that’s the tell. What I did with the time mattered more than the time itself.

The same instinct that made me want to fix her work for her is the instinct AI is quietly feeding all day long.

A couple of years ago, a team I backed walked into a room with leadership and got taken apart. Not for the work — the work was good. They’d been fully in sync with the product owner, felt they had market fit, and designed something that made the best of a complex, anti-customer policy.

Leadership asked the only question that mattered. Why hadn’t anyone pushed back on the policy itself? Why jump straight into the comfortable work of design instead of going after the actual core of the problem?

I’d backed that team through every stakeholder session. I did it on an assumption — that someone, somewhere upstream, had done the diligence. Had asked whether the policy was fixed or just unquestioned. Whether there was any wiggle room at all. Nobody had. We’d all made the best version of a premise nobody had tested.

In the room, I stepped in front of them and said it was my call. That we’d made assumptions being proved wrong in real time. That the recommendation was to go back to square one and bet we could change a deeply anti-customer policy. You can recover a project from that. What you can’t do is un-know what it costs to skip the question.

AI is that same handoff, automated. You hand it a premise and it builds the best possible version of it. It optimizes inside the prompt the way that team optimized inside the policy — brilliantly, tirelessly, and without ever once asking whether the thing it was told to do was the right thing to do.

This isn’t a senior problem. It used to take seniority to be the one who hands a premise down and never checks it. Now anyone can. The junior designer I made time for is handing premises to AI too — every prompt is a small version of that room. The tool doesn’t care how long you’ve been doing this.

It won’t ask. It can’t. The questioning of the premise — the diligence that the work is even pointed at the right problem — is the one part you can’t hand to a person or a machine. Not because they’re not capable. Because it only counts when you own it. The answer in the room can’t be “the model didn’t flag it.” The answer has to be your name.

AI doesn’t skip the making. It raises the floor on the making, and it gives you that time back clean. The temptation is to pour it straight back into more passes, more polish, more of the thing the machine already does well. That’s the comfortable space. It always was.

The diligence is the uncomfortable space. It’s slower, it’s less visible, it never photographs well, and it’s the only thing standing between a team and a room that asks the question they never did.

The time AI hands back and the part you stay answerable for are the same part. I’m not always sure I’ll spend the next twenty minutes that well. I just know which room I’m trying to stay out of — the one where the work was made well and aimed wrong.