← michele.it

The Loop.

A self-directed process that runs recursive autonomous optimizations: it reads the current state, picks the single highest-leverage move, executes it end to end, verifies the result, logs it — then feeds that new state back into itself and runs again. Every pass compounds on the last.

open the recipe
  1. 01

    Re-read state

    Start each run by re-reading every prior artifact and the last iteration's output, so it always plans from the latest truth, not stale memory.

  2. 02

    Pick the highest-leverage move

    Score what would move things most right now and choose the single best move for this pass.

  3. 03

    Execute autonomously

    Do the work end to end — research, write, build, fix, deploy — without pausing for step-by-step approval.

  4. 04

    Verify the lift

    Check its own output, keep only changes that demonstrably improve the result, discard regressions.

  5. 05

    Log and re-enter

    Write the new state, then feed it back into step one. The loop tightens each pass.

  6. ↻

    back to 01

Pass 1Pass 2Pass 3Pass 4Pass 5
Fig. 1 — successive passes along a single axis. Each compounds on the last; the dashed arc returns to re-read state and run again.
Read statePick moveExecuteVerifyLog & re-enter∞repeats
Fig. 2 — the five stages of one iteration, running clockwise, with no terminal step.

Read state, optimize, log, repeat.