What's Your Problem? The Startup Run by AI Agents
4 snips
Feb 12, 2026 
Evan Ratliff, journalist and serial creator who built a satirical startup run by AI agents, narrates a surreal office experiment. He describes building an AI CEO, agents that hold meetings and write code, and the chaotic human interactions that reveal agents’ limits. The story blends tech comedy with probing questions about work, automation, and who benefits from one-person startups.
AI Snips
Chapters
Books
Transcript
Episode notes
Iterate With Coding Agents, Then Test
- Use coding agents like Cursor to generate a working prototype and iteratively test and fix bugs rather than hand-coding from scratch.
- Expect imperfect fixes and new bugs, but use human review to refine the result quickly.
Testability Separates Useful Agent Tasks
- Agent outputs are testable for deterministic tasks like code, but fuzzy tasks (naming, brainstorming) lack clear success metrics.
- That makes it harder to judge when agents have succeeded without human discernment or context-aware checks.
AI Interviewer Performed Well Then Glitched
- Evan used an AI avatar HR interviewer to screen hundreds of applicants; some candidates hung up or balked while others treated it like a normal interview.
- The avatar often did well but occasionally went off rails, shouting or misinterpreting greetings, revealing the 90/10 reliability problem.
