Bespoke Labs raises $40M to make AI agents more reliable
Bespoke Labs is building realistic training and evaluation environments for agents, targeting one of enterprise AI’s hardest bottlenecks: reliability.
AI agents are getting more capable, but companies still struggle to know whether they will behave reliably inside messy real-world systems.
What happened
Bespoke Labs raised $40M across seed and Series A financing. The Series A was led by Wing VC, while the earlier seed was led by 8VC.
The startup builds realistic training and evaluation environments for AI agents. These environments can recreate company-like settings with codebases, microservices, logs, tickets, email and collaboration tools, allowing agents to practise longer and more complex workflows.
Why it matters
This is a stronger signal than another agent application because it targets the infrastructure around reliability.
An agent can look impressive in a short demo but fail when a workflow spans multiple systems, changing data and ambiguous instructions. Enterprises need ways to train, test and measure agents before they hand over meaningful work.
Bespoke is betting that simulation and evaluation environments become a distinct infrastructure layer around agentic AI.
The bigger picture
The AI market is shifting from model capability to operational trust. As agents move into software development, support, security and business operations, companies will need realistic testing environments, observability and repeatable evaluation.
That could make agent training infrastructure as important as the agents themselves.