Bespoke Labs Raises $40M to Build the Training Grounds for Reliable AI Agents

Bespoke Labs, a startup building the reinforcement learning environments used to train and evaluate AI agents, has raised $40 million — a bet that the scarcest resource in the agent era isn't models or chips, but the simulated workplaces where agents learn not to fail.

The Round

The bulk of the funding, $31.75 million, came via a Series A led by Wing VC, joined by Mayfield, The House Fund and employees of major AI firms including Anthropic. The company had earlier raised $8.25 million from a consortium that included Google DeepMind chief scientist Jeff Dean. Founded in 2024 by Mahesh Sathiamoorthy and Alex Dimakis, Bespoke Labs builds RL environments and infrastructure that let frontier labs and enterprises train, evaluate and improve long-horizon agents for production.

Why Environments Are the Bottleneck

The industry's core agent problem is well documented: models complete short tasks impressively but degrade over long autonomous runs — compounding small errors across the multi-hour workflows enterprises actually want automated. Post-training against realistic, gnarly environments is currently the most effective fix, which is why environment construction has quietly become one of AI's hottest categories.

The timing is instructive. Cognition's SWE-1.7, released this week, demonstrated that aggressive RL post-training on top of an open base model (Kimi K2.7) can approach frontier coding performance at a fraction of the cost — precisely the pipeline pattern Bespoke sells the picks and shovels for. As enterprises fine-tune open-weight models like GLM-5.2 and Llama rather than paying frontier API rates, demand for post-training infrastructure compounds.

The Market Context

Post-training startups now form a distinct funding category, alongside evaluation and data-curation firms, in a venture market that poured a record $510 billion into startups globally in the first half of 2026 with AI taking the dominant share. Bespoke's wager is that whichever models win, they will all need somewhere to practice — and the gym owner gets paid either way.