The hard part of AI agents isn't prompting, it's being the director

OGMYT · reddit · 2026-08-21

The author argues that the real challenge with AI agents isn't prompting, but directing them. Agents are good at work but bad at knowing when to stop, often drifting without bounded jobs, a definition of done, and clean state handoffs. The key improvement was shifting from trusting the agent's word on completion to actual verification. Agent orchestration now feels more like project management with machines than finding a magic prompt.

Original post →

More from coding & agent

coding & agent channel →