How Epsilab launched Epsilab on X
Builds RL environments for long-horizon enterprise work with real data and experts
What Epsilab launched

Epsilab is working with frontier labs to build RL environments for long-horizon enterprise work. Today, low-quality data is an industry-wide problem. We founded Epsilab to bring together real-life data and top 1% domain expertise to solve this. Backed by @ycombinator (F26). https://t.co/YbWaSybrZe
Transcript
How LaunchVideo.com rated the launch
LaunchVideo.com rating
About the launch
Epsilab builds reinforcement learning environments that frontier labs can use to train agents on enterprise workflows that stretch across many steps. Based in San Francisco and part of Y Combinator's Fall 2026 batch, the lab recreates enterprise software, tools, and complex professional tasks, then checks whether an agent finished the job. It builds those environments from real-world data and the judgment of experienced practitioners, so training tracks how the work is done inside companies. The company says low-quality training and evaluation data is already a widespread problem. On its Y Combinator page, Epsilab points to benchmarks such as TerminalBench and OSWorld, which it says include broken tasks, impossible success criteria, and graders that return false positives and false negatives. It is trying to take expert knowledge that still sits in proprietary workflows and private company data and turn it into environments where models practice that work, and where grading has to separate a finished task from a failed one. Epsilab was founded by Joanne Wang, Fengjiao Peng, Evan Hu, and Matthew Olsen. Wang, cofounder and CEO, previously built the healthcare AI startup Copper Health (YC W24) and worked in investment banking and private equity. Peng previously cofounded Shortbread (YC W23), did reinforcement learning research at the MIT Media Lab, and worked as a machine learning researcher at Vimeo. Hu was Shortbread's cofounder and CTO and did LLM research at UC Berkeley. Olsen previously worked on quantitative infrastructure at Tower Research Capital, a high-frequency trading firm.
Tags
Questions
Frequently asked questions
The Epsilab launch post reached 310K views on X, with 247 likes, 29 reposts, 102 comments, 41 bookmarks.
Epsilab launched on X on November 11, 2026 with the message "Builds RL environments for long-horizon enterprise work with real data and experts".
Epsilab builds reinforcement learning environments that frontier labs can use to train agents on enterprise workflows that stretch across many steps.
LaunchVideo.com rated the Epsilab launch 4.00 out of 5: hook 4/5, editing / creativity 4/5, copy 5/5, sentiment of launch 4/5, distribution strategy 3/5.







