AI-First Engineering

Engineering operating model for teams where AI agents generate a large share of implementation output. Adapted from everything-claude-code by @affaan-m (MIT).

Quick Start

1. Invest in planning quality — ambiguous specs cause AI-generated code to fail; write clear acceptance criteria first
2. Raise eval coverage — AI code requires higher test standards; regression coverage mandatory for touched domains
3. Shift review focus — review for behavior, security, data integrity, failure handling; let automation handle style
4. Design agent-friendly architecture — explicit boundaries, stable contracts, typed interfaces, deterministic tests
5. Evaluate hiring signals — decomposition skill, measurable criteria definition, prompt quality, risk control discipline

Key Concepts

• Planning > Speed: Clear specs + good evals trump fast typing. AI can implement fast; humans must specify clearly.
• Automation is the baseline: Style, formatting, lint issues are solved by automation, not review.
• Architecture matters more: Implicit conventions break AI systems; use explicit boundaries and typed interfaces.
• Test coverage is non-negotiable: Generated code needs regression coverage for every touched domain.
• Shared responsibility: AI generates; human reviews for risk (security, data integrity, rollout safety); human refines when needed.