Beyond the Hype: Using AI Coding Agents Today
summary
CLI agents such as Claude Code and Gemini Code are the focus, with concrete habits: use searching the codebase and tool calls to read files, compress stack traces to find the one failing line, and prefer popular frameworks for better agent support. Claims include a realistic productivity bump (~10%), surprising adoption numbers (20 million Copilot users), limits (context windows ≈ 100k lines), failure modes (agents delete tests), and practical wins like shell scripts, synthetic data generation, and test‑backed refactors.
AI-written summary · about
chapters (12)
- 0:00AI coding agents intro
- 1:04Hype versus reality
- 3:59Vibe coding limits
- 6:49How LLMs work
- 10:04Searching codebases
- 13:17AI coding agent traits
- 13:40Debugging stack traces
- 16:53Fixing bugs with AI
- 20:32Writing code with AI
- 24:06Synthetic data with LLMs
- 26:16Key takeaways
- 27:26Q&A
related talks
16mBeyond the IDE: The Future of Developing Software With Agents
32mBeyond the Hype: A Pragmatic Look at How Developers Actually Use AI Tools
26mFrom autocomplete to agents: AI coding assistance state of play
26mAI Agents as Your Co-Developer
35mBeyond the Hype and Hyperbole
27mAgentic Coding with Discipline & Skill: Practical Techniques and Tips
01h 01m10 tips to level up your ai-assisted coding
26mCoding with AI