Victor Sanh: overfitting to users is the job when building vertical agents
VictorSanh · x · 2026-11-10
HuggingFace's Victor Sanh argues many AI engineers building vertical agents think like researchers, not engineers.
- Researchers fear overfitting because generalization is the goal; users only care about inputs they send daily — narrow, repetitive ones.
- The engineer's job is making every input that matters in-distribution: overfit to the patterns users actually send (not individual samples), and pull in new breaking inputs fast.
- Accuracy on real inputs is the priority; overfitting to users is the job.
More from coding & agent
- 20 Python MCP Tools Worth Knowing, From Server Building to Gateways, Security and Eval — MaryamMiradi · 2026-11-10
- Prompting tip: describe the problem, not your solution, for better agent grilling — mattpocockuk · 2026-11-10
- The One-Person Million-Dollar AI Agency: An Agent-Stack Blueprint With the Math — boringmarketer · 2026-11-10
- Developer natively manages 25-30 parallel agent sessions on macOS with Rex — jonathan_wilke · 2026-11-10
- Pi Dash orchestrates 10 Muse Code agents with Muse Ember, rewriting 178K lines in a day — AIatMeta · 2026-11-10
- 125 prompt injections, zero executed: testing a propose-and-policy agent architecture — RI4D · 2026-11-10