Agentic Engineering in Action with Mitchell Hashimoto
AI Summary
This video is the first session on agentic engineering featuring Richard from Zed and Mitchell Hashimoto, creator of the Ghosty terminal emulator. They discuss using AI agents, specifically large language models (LLMs), to improve software quality beyond just coding speed. Mitchell shares his experience using Claude AI models to help implement and refine a new undo feature in Ghosty, involving technical challenges of working in the Zig language and Swift for MacOS. They talk about the importance of well-scoped, guided prompts to get better AI output, the limitations of current AI in architectural design and complex low-level coding, and the benefits of heavy commenting and iterative prompt refinement. Mitchell describes his workflow including running multiple AI agents simultaneously, manual code cleanup, and test-driven development practices while leveraging AI. They also explore how AI-assisted coding influences project structure and personal productivity habits. The conversation highlights practical advice for effectively integrating AI tools in software engineering projects, managing expectations of AI capabilities, and maintaining code quality and deep understanding in a high-quality, real-world project environment.