Agents that learn on the job.
Train your agents on their own work. Quality goes up, latency and cost come down, and you own the flywheel.
The harness is not enough.
Neither is the model.
Today’s agents are a frozen model with a notepad. They write down what they learn and reread it on every task, like Leonard in Memento. When the model learns inside its harness, what it writes down becomes what it knows.
- Y Combinator
- A.Capital
- SignalFire
- Microsoft M12
- Samsung Next
- Essence VC
- Clem DelangueFounder & CEO, Hugging Face
- Lasse EspeholtEx-DeepMind; founder, Ineffable Intelligence
- Charlie SonghurstBoard director, Meta
- Arash FerdowsiFounder, Dropbox
- Zehan WangHead of Applied AI Research, Twitter
- Saturnin PugnetFounding team, Worldcoin
- Bob van LuijtFounder & CEO, Weaviate
- Philip RathleCTO, Neo4j