Book a free chemistry call
Book a free chemistry call

Your org tripled overnight

Org chart: a Head of Product over five new managers (PMs and engineers), each managing three AI agents. Caption: You're tired because your org tripled.

AI agents are skip-levels we can't keep up with, and your PMs and engineers just became their managers. The fix is a familiar leadership one: coach the judgment early.

Product execs seem surprised when I offer them a reason they're so tired: their org just tripled overnight.

That's what happened when agents came onboard. These AI wunderkinds are eager but overactive, increasingly impossible to keep track of. They're brilliant, but lack judgment and context. And they're not even direct reports, they're skip-levels! Our PMs and Engineers have all been promoted to new managers.

I've been reluctant to add to the AI noise, which is partly why I haven't posted in months. But where we keep talking about tokens and context windows, I see a familiar leadership problem...

We're not coaching early enough. Or at all.

Instead I see product leaders working overtime in the weeds to catch mistakes and keep up. It reminds me of my former micro-managing days. I was cranky from sleep deprivation and wouldn't stop telling people what to do. It also didn't scale! But I micro-managed because I didn't trust their judgment and didn't yet know how to fix that.

So how do we cultivate the kind of judgment that makes the same, or better, decisions than we would make?

Ask first

Coaching a team is all about probing questions, well before anything ships. Even if we have a good answer, asking first is the better tool for sharpening judgment. Questions like:

  • What are you least sure about?
  • How well do these evals reflect the real world?
  • What's the critical pain point this new capability solves? How do you know?
  • What leading indicator will tell you this experiment is working?
  • What other (equally fast-moving) product areas does this touch?

Do pre-mortems, no matter how cheap iterating has become. Prompt your humans for evidence, from data and real user interviews. Skim the Claude artifact but then have a real debate about it. Their brain is still the quality ceiling for what gets shipped. So ask and ask and ask until you're satisfied with the answers.

And please... check, continually, that their answers align with the northstar vision and clear success metrics you set out for the team. (Because you did that. Right?)

We can't keep up with our skip-level agents, but we can coach the judgment gap in our humans who manage them.

How are you building judgment on your team right now?

This post originally appeared on LinkedIn.