Arga Labs raises $10M to build training environments for enterprise AI agents

Arga Labs announced a $10 million seed funding round in August 2026, aimed at solving one of the persistent challenges in enterprise AI: reliably training agents to navigate complex business software before they’re deployed.

The round was led by General Catalyst, with participation from Box Group, Emergence, Gradient, and SV Angel.

The San Francisco-based startup builds digital twins of enterprise software platforms — including Salesforce, Workday, and email clients — that replicate the full structure of those programs, including permission systems and web hooks. This gives developers a controllable, resettable environment in which to run the kind of large-scale reinforcement learning needed to train AI agents on real-world tasks.

CEO and co-founder Philip Li explained the problem with a practical example: an agent must determine whether a lead created in Salesforce and a separate contact reached through HubSpot represent the same company, confirm that only one email has been sent, and identify the correct recipient. “Can the agent correctly identify that these two are the same company?” Li said. “Are they able to check whether or not they’ve only sent the email once?”

Standard enterprise software makes this kind of testing difficult. Systems like Salesforce and Outlook have no built-in way to reset after each training run, making the tens of thousands of repetitions required for reinforcement learning impractical. Arga’s environments can be reset or modified on demand and run in parallel, allowing agents to be trained across multiple interacting systems simultaneously.

General Catalyst managing director Yuri Sagalov, who also runs the firm’s seed program, said the need for this type of tooling is growing. “Having a repeatable sandbox environment is very important, and much more important with agents than it was with humans,” Sagalov told TechCrunch.

Li drew a parallel to AI coding tools, which have advanced rapidly in part because software development already has mature infrastructure for deploying, reversing, and analyzing code — infrastructure that makes reinforcement learning environments easier to build. Similar tooling for business software has largely been absent, and Arga is positioning itself to fill that gap.