"The 80% Problem: Where AI Leaves the Engineer Behind"

A thoughtful essay making the rounds today by Jonathan Beard digs into something every developer has felt but few have named: AI coding tools get you to eighty percent completion with startling speed, but the last twenty percent — the edge cases, the failure modes, the operational hardening — was always where the engineer actually lived.

The fluent surface

The argument is precise. Ask a model to build a feature and it produces the happy-path logic, the passing test, the scaffolding that runs in a demo. What it quietly skips are the things that only emerge from operational experience: the idempotency key for racing requests, the backoff and jitter that prevent a retry stampede, the structured log that makes a 3am outage diagnosable. Every one of these is invisible during development because the dev environment never exercises the condition that would expose it. The code compiles. The test passes. The demo works. Then it meets concurrency or a traffic spike, and it isn't done at all.

Synthetic competence

Beard coins the term "synthetic competence" — output with the surface texture of understanding and none of it underneath. A developer with AI can produce fluent design docs, plausible code, and confident architectural recommendations while the depth behind them may not move an inch. The old tell — not understanding something showed up as not being able to build it — is gone. Now the building is cheap, and understanding has to be checked some other way: by whether you can predict how the thing fails and name the assumptions it's making.

The ironies of automation

The piece draws heavily on Lisanne Bainbridge's 1983 paper "Ironies of Automation," which warned that the more reliable the automation, the more crucial the human becomes in the moments it fails — and the supervisory role is precisely the one that least prepares you to step in. Automate the routine work and you've removed the practice that kept the operator's skill alive. Today's senior engineers earned their judgment by grinding through edge cases when that was simply the job. The next cohort walks into a workplace where the boilerplate is already automated away, and the foundation under the supervision may retire with the people who built it.

The takeaway

None of this is an argument against using AI — the speed is real and valuable. The fix, Beard suggests, is deliberate: keep the apprenticeship, defend the on-call rotation against the efficiency that wants to automate it away, and treat the moments of productive struggle as the only curriculum that's ever produced someone who can hold the last twenty percent when the first eighty leaks into it.

Read the full essay at jonathanbeard.io.