Peter Elias

Peter Elias is founder of Probably, a company building a verifiable data agent backed by a $9 million Andreessen Horowitz seed. In TechCrunch coverage he describes the product as a data-science mech suit: LLMs draft, deterministic validators bounce failures, and smaller local models become viable.

Harness over ever-larger models

Elias's public argument is that labs are incentivized to sell more correction loops, while Probably invests in context and validation so the model does less guessing. That leads to local, quieter models for precision work.

Data science first, then other precision jobs

The first product targets analysts over tables and warehouses. Elias has said the same engine can extend to accounting or medical services-any precision-sensitive use case-once the harness pattern holds.