1- UX > DX: Choose the tech stack that delivers the best performance and native feel for the user. Invest the time saved by AI into maintaining separate, high-quality native codebases rather than compromising with cross-platform wrappers.
2- Security > Everything Else: If there is more than one solution, choose the most secure one, even if it slightly impacts usability.
3- The Truth Chain: [Spec → Test → Code → Verify → Commit] ↺
Treat the Spec and Tests as the 'Truth'. If the AI implementation passes the tests but fails the spec, the implementation is wrong. We don't write code; we solve problems. The code is just the paperwork the AI files to prove the problem is solved.
Avoid the perfectionism trap. Use the speed of AI to run through the Truth Chain multiple times. Clarity happens through iteration, not just contemplation.
4- Automate Security: AI scans for anomalies; SAST scans for rules in the verification stage.
5- Docs-as-Code: Feature requests are not messages; they are Pull Requests to the spec documentation.