Xiangyi Li
Founder
Evals and environments for agents · creator of SkillsBench and PostTrain.com
About BenchFlow
A frontier environment lab for AI agents.
We build the environments AI agents need to learn — and be evaluated on — real computer work, not static prompts.
People
A small team, close to the work.
We work across agent research, evaluation, systems, and learning — from the Bay Area.
What we build
Real workplaces, not static prompts.
BenchFlow is a data lab. We build stateful environments with services, files, tools, verifiers, traces, and replay, so agents can learn from whole workflows.
We ship SkillsBench for procedural skills, ClawsBench for high-fidelity simulated workplaces, and BenchFlow, the runtime that runs both.
Built with the community.
BenchFlow started in late 2024. We work with frontier model labs, the agent skills community, and academic partners through Agent Skills ’26 at CAIS and the May 2026 SkillsBench 1.0 launch.
Work with us