Tip-offs
•
Each week, we surface promising startups & founders through data signals.
This week, we're spotlighting RL Environments →
Labs are buying practice, not data
Frontier labs have a new scarce input, and startups are racing to sell it.
Pre-training gains are flattening. The models improving fastest are the ones that practice - agents dropped into simulated codebases, browsers, and CRMs, graded on thousands of attempts until they stop failing. Those practice worlds are called RL environments, and building them well is hard, slow, expert work. Labs that once bought labeled data now buy environments.
RL Environment Startups
Seed-stage companies building the simulated worlds where AI agents learn to work, ranked by website visit growth over the past 3 months.
Halluminate | 429% growth in 3 months
Managed sandboxes and datasets for training computer- and browser-use agents. Raised a $500K Pre-Seed in Sep 2025.
San Francisco, United States
Vmax | 309% growth in 3 months
Automates converting proprietary data into RL environments for LLM agents. Pre-Seed; funding undisclosed.
San Francisco, United States
Andon Labs | 308% growth in 3 months
Benchmarks and evaluations for long-horizon AI agent safety and coherence. Raised a $500K Pre-Seed in Apr 2024.
San Francisco, United States
BenchFlow | 124% growth in 3 months
Open-source evaluation infrastructure and benchmark hub for AI agents. Raised a $1M Pre-Seed in Feb 2025.
San Francisco, United States
Mechanize | 48% growth in 3 months
High-fidelity RL environments and evals for frontier coding agents. Raised a $9.1M Seed in Apr 2026.
San Francisco, United States
RL Environment Founders
Founders building the gyms where AI agents learn to work.
Justus Mattern
Co-founded Revideo (YC S23). Led RL research at Prime Intellect. Now co-founder of Proximal, RL environments and benchmarks grounded in real codebases. Raised a Seed in 2026.
San Francisco, United States
Nicolai Ouporov
Founding engineer at Respell (acquired by Salesforce, Jan 2024). Now founder & CEO of Fleet, RL gyms replicating enterprise software — ~$1M to ~$60M run-rate in four months. Raised a $15M Seed.
New York, United States
Pranav Putta
Georgia Tech. Previously at MultiOn and Tonic.ai. Now co-founder & CTO of Plato, simulated worlds for training and evaluating browser agents. Pre-Seed; funding undisclosed.
San Francisco, United States
Spencer Mateega
Wharton (BS), Penn (MS, Computer Science). Now co-founder & CEO of AfterQuery, expert-built datasets and RL environments for AI labs. Raised a $30M Series A in Apr 2026.
San Francisco, United States
Christopher Settles
UIUC (Computer Science). Machine learning at Uber AI. Now co-founder & CEO of Refresh, training gyms for coding and computer-use agents. Raised a Pre-Seed in 2025.
San Francisco, United States
Carried Interest
Articles & podcasts of the week
