Fragments January 8

Takeaways

• Anthropic internal report: AI used for 59% of dev work, ~50% productivity lift; debugging and understanding code dominate, new-feature implementation rising.
• ~14% “power users” see much larger gains; AI helps devs work outside core expertise—raises career/profession concerns.
• Obie Fernandez case study: ~13k LOC production app (auth, semantic search, MCP server, webhooks, tests) built over holidays with Claude Code.
• TDD as forcing function in AI-assisted dev—red-green-refactor keeps human in loop when directing massive codegen.
• Martin Fowler fragment curates links; value is in workflow granularity (what humans review, correct, decide), not hype.

Notes

Supports tight human loops + verification in agentic coding manifesto and Claude Code practices—concrete data point for career narrative on AI-augmented engineering.

Open questions

• Can vault agent workflows adopt TDD-like checkpoints (dry-run skills, diff review) analogous to red-green-refactor?