Arga raises $10M to build enterprise AI training environments

The company builds full-scale digital twins of enterprise software such as Salesforce, Workday, and email clients. Unlike typical stateless API testing environments, Arga clones the entire software, including permission systems and web hooks, allowing agents to be trained across multiple systems in a realistic and resettable setting.

Arga’s CEO and co-founder Philip Li illustrated the need with a scenario where an agent must handle a lead created in Salesforce while a colleague separately contacts the same company through HubSpot. The agent must determine if the two entries refer to the same company, avoid duplicate emails, and decide which sales opportunity to address. Li said agentic systems still struggle with such ambiguity, and his company’s tools aim to close that gap.

Training agents for complex enterprise tasks usually relies on reinforcement learning, which requires running scenarios tens of thousands of times. But enterprise software like Salesforce or Outlook cannot easily be reset or cloned, making such massive-scale testing nearly impossible. Arga’s digital recreations, designed like crash test dummies for software, can be reset or modified on demand. The company can also run many environments simultaneously to train agents on interactions between different programs, replicating a person’s full work environment with overlapping tasks across systems.

This approach is intended to close what the company describes as a “reinforcement gap” between coding and other applications. AI coding tools advanced quickly because developers already had sophisticated deployment, reversal, and analysis tools that make it easy to set up reinforcement learning environments for code. Similar tools do not yet exist for most business software, but Arga believes creating them will make AI agents significantly more capable and potentially transform industries the way they have transformed coding.

General Catalyst’s managing director Yuri Sagalov, who runs the firm’s seed investing program, said he sees a growing need for agentic testing tools like Arga. He noted that much of the economic value from agents will come from using business applications, and that having a repeatable sandbox environment is far more important for agents than it was for humans.