Bespoke Labs raises $40M Series A to build training grounds for reliable AI agents
What's the deal? Bespoke LabsDealroom has a profile for this one. Try Dealroom →, an AI agent training platform, has raised $40 million across a Series A led by Wing Venture CapitalDealroom has a profile for this one. Try Dealroom → and a seed round led by 8VC. Participants included Mayfield, The House FundDealroom has a profile for this one. Try Dealroom →, dbt Labs chief executive officer Tristan Handy, Jeff Dean, Resolve AI chief executive officer Spiros Xanthos, DevRev chief executive officer Dheeraj Pandey, and angels from Anthropic, OpenAI, and Meta.
What's the endgame? The company builds the environments and tooling that sit underneath reliable agents — sandboxes that mimic real companies, with large codebases, microservices, logs, tickets, email, and Slack. The goal is to let agents learn long-horizon workflows and be measured against them.
Why now? Today's agents can write code and complete short tasks but still struggle to work autonomously over hours or days. Independent benchmarks from METR show the length of tasks agents can reliably complete is doubling roughly every seven months — a pace that demands environments of matching complexity.
What's the money for? Bespoke Labs will expand its research team, scale its environment-building infrastructure, and accelerate business development. It takes a research-first approach staffed by scientists and engineers rather than contractors.
The company is a core contributor to Terminal-Bench, a widely cited benchmark for agentic capability, and the team behind OpenThoughts, an open reasoning dataset downloaded more than 500,000 times and used by groups including Meta and Amazon.
In their words: "Frontier labs, enterprises, and all organizations relying on reliable agents need access to high-quality environments," said co-founder and chief executive officer Mahesh Sathiamoorthy. "This is the critical piece needed to optimize and develop agents."
The signal: The round ranks in the 94th percentile for US Series A deals in its industry, a sign that investors are backing infrastructure for agent reliability and evaluation — not just frontier models or agent front-ends. That points to a maturing market where enterprises care about safety, robustness, and measurable performance, and where agent training environments are emerging as a category of their own.
Read more: business.theeveningleader.com
