Arga Labs announced a $10 million seed funding round on Wednesday, led by General Catalyst with participation from Box Group, Emergence, Gradient and SV Angel. The startup develops test and training environments designed to make agentic AI work reliably inside enterprise applications.
The problem Arga is addressing
Enterprise systems such as Salesforce, Workday and email clients maintain complex state, permissioning and integrations that are difficult to reproduce in traditional test setups. While many testing environments expose stateless API endpoints, Arga Labs creates full-scale digital twins of target applications — replicating structure, permissions and web hooks so the whole program behaves like the real system.
Having complete control over the environment makes it simple to reset or modify, and allows running many instances in parallel. That lets teams train agents on the complicated interactions that occur across multiple programs and knowledge systems, simulating overlapping, real-world workflows.
A practical example
Philip Li, CEO and co-founder of Arga Labs, gives a common scenario: one salesperson creates a lead in Salesforce while a colleague separately reaches out via HubSpot. Questions arise: can the agent recognize these are the same company? Can it check whether an email has already been sent? Which opportunity should receive the outreach?
Agentic systems still struggle with these kinds of ambiguities, and Li sees Arga’s tooling as essential for improving those capabilities.
Why standard reinforcement learning falls short
In principle an agent could be trained via reinforcement learning (RL) by running a scenario tens of thousands of times and selecting only successful strategies. In practice, enterprise software makes that scale of testing nearly impossible: there is no easy way to reset or clone a system like Salesforce or Outlook for repeated RL runs.
Arga’s solution is to digitally recreate the software — a ‘crash-test-dummy’ style replica of the application — giving engineers full control. The company can reset environments, change conditions, and run many simultaneous instances to train agents on cross-application workflows.
Industry implications
Arga aims to close the reinforcement gap between coding tasks, where sophisticated RL tooling already exists, and business applications, where such tools are still lacking. The availability of repeatable, controllable sandboxes has accelerated RL progress in coding; Arga’s approach seeks to bring similar advances to enterprise apps, which could markedly improve how agents perform inside businesses.
Yuri Sagalov, managing director at General Catalyst and head of the firm’s seed program, told TechCrunch he sees growing demand for agent-testing tools. “I think that a lot of the economic value from agents is from using business applications,” Sagalov said. “Having a repeatable sandbox environment is very important, and much more important with agents than it was with humans.”
Conclusion
With $10 million in seed funding, Arga Labs is building full-scale digital twins of enterprise software to enable scalable, repeatable training of AI agents. By making it feasible to reset, modify and parallelize environments, the company hopes to accelerate agent capabilities on real-world, multi-system business workflows.
