AI research & engineering · Chief Architect, TextQL
Gabriel Tomitsuka
I build the environments language-model agents are evaluated in, and the infrastructure they run on.
I'm employee #1 at TextQL, now a Series A company of about fifty people. I built Ana, one of the first public AI agents, and as Chief Architect I lead the systems agents run on: Sandcastles, sandboxes on Kubernetes that start in under 300 ms and scale from 10 to more than 1,000 in three minutes, and Ontology, a git-versioned semantic layer that agents read first and write back to.
Running agents at that scale took me upstream. I fixed a three-year-old concurrency bug in Docker Engine's API client, the first fix in its 27.0 release notes, and contributed to Apache Arrow's Go library. Before TextQL, in 2020, I founded a startup that connected people through books.
Latest paper · arXiv 2610.02122 · October 2026
Argo-Bench: Evaluating Data Agents on Enterprise-Scale Workflows
An evaluation framework for measuring data agents on their decision-making in a high-fidelity simulated marketplace. Agents make the calls that carry real enterprise value, such as allocating budgets, in a New York City food-delivery marketplace simulated at true scale, with 81 million orders, and each action is graded by its consequences in the simulator.