🤖 AI as a pair programmer: What works, what doesn't?

[copyleftdev]

This entire repo was created through human-AI collaboration using Windsurf/Cascade.

What the AI did well:

• Generating consistent AsyncAPI schemas
• Transforming RFCs to machine-readable JSON
• Writing automation scripts (issue importer)
• Creating diagrams as code
• Maintaining consistency across 8 RFCs

What required human judgment:

• Architectural decisions (TigerBeetle-style vs microservices)
• Domain modeling (what events matter)
• Acceptance criteria (what "done" means)
• Priority and sequencing

The key insight

The structured artifacts (specs, RFCs, JSON schemas) become a persistent context that AI can use across sessions. GitHub Issues become a state machine for tracking progress.

Questions

• How are you using AI in your development workflow?
• What tasks do you find AI excels at vs struggles with?
• Do you structure your repos to be "AI-friendly"?

Let's share what's working! 🧠