FrameworkLevel 6 · ConcentrationDrift & repair

Drift Is The Default

Everything that generates, drifts. Correctness is not something you install once — it's something you maintain by returning, and the interval is the quality.

The shape

Any system that generates — an AI agent, a creative human mind, a working team — drifts by default. Not because it's incapable, but because it's stochastic: it wanders off the thing it was doing, a little each step, and twenty steps later it's confidently solving a different problem than the one you gave it. Each individual step looked fine. The trajectory rotted.

The consequence: correctness is not something you install once. It's something you maintain by returning. You externalize the goal, periodically check current activity against it, catch the drift early, and steer back. The real performance number isn't "how good is each step" — it's how short the interval is between drifting and noticing. This is exactly concentration in the meditative sense: not the absence of wandering, but catching the wander sooner each time until the return becomes continuous.

And the variability that causes drift is the same variability that produces novelty. You don't want to eliminate it (that kills the engine) — you want to ride it, leaning into the drift early so the trajectory stays on target.

Why it matters

Long-running work fails in a way that's invisible until it's expensive. The session, the project, the agent run all look productive the whole time — and then you look up and ask "what did we actually accomplish?" Nobody made a dramatic mistake. There just was never anything checking, on a short enough interval, whether all that motion still served the goal. The drift compounded silently.

This kills the most common false belief about AI (and about your own focus): that if you set it up right, it'll stay on track. It won't. Nothing stays on track on its own. The teams, agents, and people who appear to have superhuman focus are not drifting less — they're returning faster, because something is checking alignment often enough to catch the wander while it's still small.

How to think with it

1. Assume drift. For any extended piece of work, the question is not "will it drift?" but "how will I catch the drift early?"
2. Externalize the goal so it can be checked against. A goal that lives only in your head morphs and softens until there's nothing to compare current activity to.
3. Make the check honest. "Are we still on mission?" → "yep" → next, is theater. The real check forces specifics: the goal is X, current activity is Y, Y serves X because Z. If you can't say Z, you just caught the drift.
4. Shorten the interval. Check every ten minutes and you catch ten-minute drifts; check every four hours and you lose half a day. The frequency of the return is the quality of the work.

Honest caveat. What this doesn't cover: the structures that perform the return without you, so an unsupervised run stays on goal for hours instead of degenerating.

Distilled from: mission control · l6 concentration · concentration engineering · seam repair · meta cognitive awareness · hiel