正文 Markdown
Wrapped up the 3-day AI Engineer conference in London. Key takeaways: Planning & verification > implementation. Humans define what to build and validate it works. AI handles the coding. The harness matters. Terminal bench showed wrapping/orchestrating an LLM changes output quality dramatically. (claude code was one of the worst performing harnesses!) Parallel multi-agent systems aren't common yet. Factory (Droid) runs parallel agents in read mode only — writes are serialized for testability. (they should try mcp_agent_mail (https://t.co/pubsV51cs1) cc @doodlestein ) Bottom line: everyone's still figuring this out. No one's cracked consistently high-quality AI-assisted software yet — and that's because these tools are evolving incredibly fast. A big thank you to @swyx and the rest of the @aiDotEngineer team for organizing this event!