VIBEKODED-OS // v1.0

// track · 9 lessons

Make the bad state impossible

If it must never happen, don't ask the model to be careful. Build it so it can't.

  1. 01Make the bad state impossibleIf it must never happen, don't ask the model to be careful. Build it so it can't.
  2. 02Don't ask, enforceFor anything that must never happen, don't ask the model to be careful. Build it so it can't happen. Asking lowers odds; structure removes them.
  3. 03The 99% problemNinety-nine percent right is a countdown, not a margin. Probability compounds across runs, so catastrophic paths need determinism, not odds.
  4. 04Make illegal states unrepresentableDon't validate against bad states. Build structures where a bad state cannot be constructed at all. Unrepresentable beats caught every time.
  5. 05Deterministic core, AI amplifierPut the must-be-exact work in deterministic code and let the AI amplify around it, never compute it. Variance helps until answers can't flex.
  6. 06Least privilege, or the blast-radius questionGrant the least privilege that does the job. Every permission is a liability if it leaks, and that gap is the difference between a shrug and a disaster.
  7. 07Safety by structure, not by promptA guardrail in the prompt is a suggestion the model can reinterpret. Put the constraints that matter in the layers the model cannot touch.
  8. 08Validation gates, or rejecting the bad state before it landsPut a deterministic, fail-closed gate between the model and anything it can damage. A check that rejects on doubt makes a fallible generator safe.
  9. 09Layered enforcement, or defense in depthNo single guardrail is enough. Stack layers that each assume the others might fail. Impossibility comes from defense in depth, not one wall.