First attempt will be 95% garbage: 6 weeks with Claude Code
8 days ago
- #Workflow Optimization
- #AI in Development
- #Code Review
- Vincent Quigley, a Staff Software Engineer at Sanity, shares his experience integrating AI into his development workflow.
- AI now writes 80% of his initial code implementations, allowing him to focus on architecture, review, and managing multiple development threads.
- He treats AI like a 'junior developer who doesn't learn,' emphasizing iterative feedback and review processes.
- His workflow involves using Claude Code for full AI delegation, with multiple instances running in parallel like a small team.
- Key strategies include creating project-specific context files and integrating AI with tools like Slack and GitHub for better efficiency.
- AI-assisted coding has evolved his code review process, making him more critical and less emotionally attached to the code.
- Current limitations include AI's inability to retain learning between sessions and confidently writing broken code.
- Despite challenges, the ROI is significant, with AI enabling faster, better solutions and handling small tasks autonomously.
- Advice for technical leaders: start small, give AI three attempts at a well-defined feature, and review the output critically.
- The future of development is about leveraging AI to work faster and create better solutions, not replacing developers.