How I Tamed Claude
summary
A practical process for getting production code from Claude Code: use OpenSpec’s explore/propose/apply/archive flow, split agents into orchestrator/worker/reviewer with no actor approves its own code, and insist on one commit per task group and a strict definition of done (build clean, tests green, coverage target). Tools used: Claude Opus/Sonnet, CMX terminal, OpenSpec, Graphify, Context Mode, Context7, GitHub. Examples: tens of thousands of C# lines generated, a change ran 111 tools, used ~177,000 tokens, produced 1,183 passing tests and a working repo.
AI-written summary · about
chapters (14)
- 0:00Welcome
- 1:59Why Claude for coding
- 3:26Claude's models and context
- 7:37Mindful engineering with AI
- 9:03OpenSpec workflow
- 15:35Separation of powers in agents
- 21:55Agent and human roles
- 26:50Limits and definition of done
- 31:04Human oversight importance
- 36:06Curating context for tokens
- 42:24Live demo: agent workflow
- 51:09Open source tools
- 55:04Five principles for AI engineering
- 57:18Q&A
related talks
01h 06mHow I Tamed Claude
44mClaude Code for Spring Developers
01h 06mHow Claude Code Works
01h 37mFull Walkthrough: Workflow for AI Coding
21mHow Anthropic Uses Claude
16mBeyond the IDE: The Future of Developing Software With Agents
47mClaude Code at Scale: An Android Deep Dive
24mClaude Plays Worth Stealing | Claude Code & AI Coding Tips