-- Backwards Build: Start with business slides → PRD → documentation → tests → THEN code. Generate code only after product makes human sense.
-- Leverage MCP: Use Model Context Protocol to connect your AI with codebases, documentation and external tools in a standardized way.
-- Standardize Aggressively: Minimal invention, heavy use of TypeScript(no to JSX)/strong typing in Python. Follow existing conventions.
-- Strict Linting: Ruff for Python, ESLint for frontend - no exceptions.
-- Verification Gates: Require AI to produce tests, docs, and diagrams with mandatory human review before proceeding.
-- Size Limits: 50 lines max per function, 200 lines max per file.
-- All in sync: Keep the entire chain -- docs, slides, code, tests -- in sync on a daily basis
-- Collaborative Review: Human-AI joint code reviews with clear quality criteria.
What other AI coding practices have improved your team's output?