Self‑Automator: a mode where AI handles both WHAT and HOW, leaving the human to step aside.
The machine runs, the rider has dismounted.
What It Is
The third mode alongside Centaur Work and Cyborg. In Centaur the human steers WHAT and HOW, in Cyborg the human gives up the HOW and keeps the direction. The Self-Automator gives up both: WHAT and HOW move to the AI, the human steps aside.
At 27% it was no fringe case in the study - and that group delivered the weaker results, even though everyone had the same tools. The difference isn't the tool, it's who holds the direction.
Important: self-automation isn't inherently bad. Deep inside the Jagged Frontier, on routine work the AI handles reliably, it's legitimate efficiency. The line doesn't run between "automate yes/no", it runs between deliberate delegation and thoughtless release.
How To Spot It
- Output gets adopted without anyone setting the direction or checking the result
- "Just let the AI do it" - and nobody notices when it solves the wrong problem
- The person doesn't get smarter while working: no checking, no steering, no learning
- The opposite: deliberate delegation that knows the task's frontier position and spot-checks
What To Do (FL1 - Team Level)
- Check the frontier position: inside the Jagged Frontier you may automate, near the edge you need the human on top
- Keep or hand over the WHAT deliberately: both are legitimate, as long as it's a decision and not a reflex
- Spot-checks instead of blind flight: even delegated work needs control cuts, otherwise you notice the drift only once it's expensive
- Keep the AI as a teacher: if you only sign off instead of thinking along, you unlearn how to check (Cognitive Load Shift)
The Trap
Handing over the WHAT without thinking. Fast, polished, without depth and without learning. For one person that's a missed learning effect. Scaled across the whole organization it stays stuck in old patterns, gets faster output but no new smarts, and tips toward Closed Door: when nobody holds the direction anymore, the org stops responding to what the AI actually makes possible.