Agentic Engineering Patterns: Code Is Cheap

• Source: https://simonwillison.net/guides/agentic-engineering-patterns/code-is-cheap/
• Author: Simon Willison
• Added: 2026-03-01
• Topic: Context engineering, agentic engineering, developer workflow

Why this matters

This piece reframes a core operating principle for AI-assisted engineering: generating code is cheap, but validating behavior, preserving context quality, and maintaining system reliability are the real constraints.

Key takeaways

1. Optimize for evaluation and feedback loops, not code volume.
2. Treat generated artifacts as disposable unless they pass strong checks.
3. Prefer workflows that produce inspectable intermediate outputs.
4. Use agents to generate explanatory artifacts when complexity rises.
5. Keep context tight, explicit, and test-driven to avoid drift.

Practical application notes

• Add executable evals to every agentic workflow.
• Require failure-path tests before merge.
• Bias toward reproducible traces over clever prompts.
• Keep prompts/contracts versioned alongside code.

Related notes

• agent-reliability-chasm.md
• closed-loop-telemetry-driven-optimization.md
• system-design-and-invariants-pattern.md