Encoding organizational policy as running software rather than written edicts, so that compliance emerges from the environment rather than obedience.
## Core Framework
When an organization cannot or will not mandate engineering behavior top-down (Netflix's "highly aligned, loosely coupled" culture), you can instead:
1. **Identify the failure mode** you need everyone to handle (e.g., vanishing cloud instances)
2. **Build software that reproduces it continuously** during working hours (Chaos Monkey)
3. **Let engineers solve it however they want** — the environment creates the constraint, not management
The key insight: this transforms a policy problem ("everyone should write resilient code") into an engineering problem ("my service keeps going down every day"). Engineers solve engineering problems.
## Why This Works
- Removes the communication bottleneck of mandating practices across large orgs
- Respects autonomy — engineers choose the solution, not just execute orders
- Acts as continuous regression testing — no drift back into fragile patterns
- Makes the cost of non-compliance immediate and personal (your service goes down)
## Cross-Domain Applications
**Organizational Design**: Any "highly aligned, loosely coupled" org faces the same challenge — you can't mandate behavior but need alignment on outcomes. Code-as-policy is one solution pattern.
**Incentive Design**: Parallels Goodhart's Law inversions — instead of measuring compliance (which gets gamed), change the environment so the desired behavior is the path of least resistance.
**Education**: Self-directed learning environments that present problems rather than prescribe solutions (Montessori, unschooling) follow the same pattern — structure the environment, not the behavior.
**DevOps/SRE**: Error budgets and SLO policies encode reliability principles as automated gates rather than team mandates. Same philosophy, different implementation.
## Key Quote
> "Chaos Monkey is a management principle instantiated in running code."
> — Casey Rosenthal, *Chaos Engineering* (O'Reilly, 2020)