The Many Claudes of My Code
6 hours ago
- The author initially doubted AI's coding abilities, treating it like Stack Overflow and manually reviewing code to maintain personal style.
- Claude Code demonstrated AI's ability to handle maintenance tasks and reduce the need to learn specific frameworks like Laravel.
- Newer AI models (e.g., Fable, GPT 5.6-Sol) now outperform the author in 80% of coding tasks, leading to acceptance and excitement about productivity gains.
- AI's fallibility is mitigated by deterministic tools that validate code, enabling self-checking with speed and sustained attention.
- The author previously delegated grunt work to a human remote developer (Hercules), which helped reframe AI as a team member rather than a threat.
- The author uses fast, less reasoning-heavy AI models (like Sonnet, 5.6-Luna) as 'Coders' for boilerplate and low-creativity tasks.
- Sophisticated AI models (like Fable/Sol, Opus/Terra) act as 'Reviewers' in adversarial testing, finding bugs and passing issues back to Coders until resolved.
- An 'Assistant/Orchestrator' (e.g., Codex CLI with 5.6-Sol) delegates maintenance, research, and coding tasks, persisting for large, one-shot tasks.
- More AI agents can lead to hallucinations, review loops, and wasted tokens; balance is key, and professional developers must read and understand code to avoid responsibility for failures.
- Skill atrophy is a real risk; vibe coding without deep understanding can lead to losing industry relevance, so AI should accelerate existing skills rather than replace them.
- Controlling the project structure and building with intention makes AI more effective, while aimless use wastes money.
- The author emphasizes that AI is a tool to accelerate human work, not replace human judgment and responsibility.