Remi Louf's talk: skip agent frameworks — hash your prompts, use structured outputs

remilouf · x · 2026-08-27

tsudo recommends Remi Louf's talk "Agent Frameworks Considered Harmful". Key ideas: instead of leaning on heavy agent frameworks, use hashing to build prompts (cacheable, reproducible, deduplicable) and leverage structured outputs to make agent workflows more reliable. The talk questions the value of framework abstraction layers.

Original post →

More from coding & agent

coding & agent channel →