Monokit is on sale: the app screens most kits leave out · See Monokit → · Read the post →

← Studio

AI adoption · 2026

Automation Solves Output. It Doesn't Solve Outcomes.

I drew two charts to separate two claims that get flattened into one: more automation always means more output, but outcomes peak and then fall. Then I watched my own audit agent sit on the wrong side of the second chart for fifteen mornings straight.

More automation always means more output. Nobody argues with that part: plot automation against output and the line only goes up.

What rarely makes the same slide is what happens to outcomes. So I drew two charts, to keep the two claims apart.

Two charts: automation increasing output while reducing human intervention, and outcomes forming an inverted U against outputTwo charts: automation increasing output while reducing human intervention, and outcomes forming an inverted U against output
Output climbs with automation. Outcomes don't.

The first chart is the pitch

Output goes up, human intervention goes down, and the two lines cross at a tidy "optimal balance". Put in less of yourself, get more out. That's the automation pitch in one picture, and it's true as far as it goes.

It's also the only chart most AI adoption decks show.

The second chart is the one that matters

Outcomes (impact, quality, whether the thing actually lands) don't climb with output. They rise, peak, then fall.

Too little output and you've under-delivered. Too much and you get volume without judgment: quality drops and the work drifts away from what anyone asked for. The best outcomes sit in the middle, and getting there takes the right amount of oversight, not the least.

I've been on the far side of that curve

My audit agent, Ledger, once filed the same issue fifteen mornings in a row against a test PR I'd forgotten to close. Correct every time. Useful zero times.

Nothing was broken. The system was running at full automation with nobody positioned to ask whether it should still be running. That's the right-hand tail.

"Optimal balance" isn't a setting

It's a point that moves with the cost of being wrong. Ledger can run almost unsupervised, because a bad issue costs someone thirty seconds to close. A pipeline that ships to production, or drafts something with my name on it, needs a lot more of me in the loop.

The chart doesn't give you a ratio. It shows you the shape of the trade-off you're managing.

I said this back in 2023, before most of this existed: AI is a lever, and a lever needs a skilled operator. Three years and one autonomous pipeline later, I'd add one thing. The operator's job isn't to slow the lever down. It's to notice when it has stopped pointing at anything worth pulling.