I manage bots better than I ever managed people

I manage bots better than I ever managed people. Which is depressing.

I have bots for email, bots for research, bots for code, bots for reviews, bots for personal stuff. Bots for everything. My ADHD and introvert brain loves it. I can jump between ten of them in an hour and nobody minds. I can manage them at 3am without getting on a call. When one of them screws up there’s no awkward conversation. I just rewrite the skill and we never talk about it again.

I’ve been reading Ben Horowitz’s The Hard Thing About Hard Things, and a lot of the lines I highlighted are about people, but they hit harder for bots.

If you don’t train your people, you establish no basis for performance management.

That’s a skill file. If I haven’t written down what good looks like, I can’t be mad when the bot doesn’t do it. I still get mad anyway. Genuinely annoyed at a bot like it’s a person.

giving the team a task that it cannot possibly perform is called crippling the army.

That’s every over-scoped ticket I’ve handed an agent and then blamed the agent for.

Here’s the part I don’t love. A researcher at Boston University had 1,261 managers review drafts with planted errors. At companies that already put AI on their org chart, managers caught 17% fewer errors when the draft was labelled as coming from an “AI employee” instead of an AI tool.

That’s me. I trust the bots I like more than I should.

Andy Grove has a line in High Output Management I keep coming back to.

delegation without follow-through is abdication.

So the bots are on a trust ladder now, same as my code. A bot earns more autonomy by not screwing up for a while. When one does screw up, I add a verification step so that mistake can’t happen again. Billing is the bottom of the ladder. I still review everything there because it touches real money.

Katie Parrott put it better than I can in this piece on AI autopilot:

The feeling of having reviewed is easy to produce. The act of reviewing is harder.

But I’ve been telling myself managing bots is easier than managing people, and it’s not. I’m just avoiding the human part. Bots don’t need me to be good at hard conversations, so I’ve stopped practicing them.

I don’t know what to do about that yet. That’s kind of why I’m writing this.

I wrote about my bots being very good at the wrong stuff and about how the constraint is me a couple of weeks ago. This feels like the same problem from a different side.

~ aq