Altitude isn't a bigger prompt

Transcript

Rethink your priors.

This is part two of four in my mini series within a series on Felix Rieseberg's post, How to be more ambitious, a tutorial for humans with models.

Last time was depth. This one is altitude.

Altitude means going up in abstraction. You stop handing the model tasks and start handing it outcomes.

Felix puts it as a question: what do you want to happen next? What's the goal beyond the goal?

His own example is a good one. At the beginning of last year, he was still asking Claude to fix individual crashes. Eventually, he asked Claude to simply keep their apps crash-free.

He's honest about the limits, too. Altitude works best when you really are after an outcome, and less well for creative work.

And, Claude, go make money, doesn't work for most businesses on its own. Not yet.

The part I like most is the plumbing. To let Claude keep the apps crash-free, he first had to make sure it could fetch crash reports, map each app version to its git commit, and get access to Windows, macOS and Linux computers, so it could reproduce and verify fixes on its own.

So here's mine. On October 21st I handed all the work on my app, Clueless, to a coding agent with one goal: take the Muse connector live for every subscriber.

It owns its pull requests and its merges. I only approve the gated steps. The auth plan, store submissions, policy text, campaign sends.

But the plumbing came first. Every bug that gets past us becomes a ticket to fix the system, not the app, with a baseline, a target, and proof required to close it.

Once, some text got clipped two points from the edge of the screen. An agent's layout audit missed it. A human caught it.

So we changed the gate to catch that whole class of bug. Not the bug. The class.

Then there's a Linear and pull request sync three times every weekday, a daily ship check, and merge gates.

Very glamorous. Nobody has ever asked me about my merge gates at a party.

That's the real work. Altitude isn't a bigger prompt. It's giving your agent the access and the feedback loop to own an outcome, so you stop feeding it task after task.

Next time, range.

Anyway. Start from the outcome. Or don't. I'm not your manager. Follow for more.