Arga, a startup building training environments for enterprise AI agents, announced Wednesday that it raised $10 million in a seed funding round led by General Catalyst. Participants in the round included Box Group, Emergence, Gradient, and SV Angel. The company鈥檚 tools are designed to address a growing problem for businesses: testing and training AI agents on complex enterprise software before deployment.
Arga creates full-scale digital twins of programs such as Salesforce, Workday, and email clients. Unlike typical testing environments that offer a simple stateless API endpoint, Arga replicates the entire software, including its permission systems and web hooks. This approach allows the company to train agents across multiple systems simultaneously, offering a more robust way to handle the ambiguities of real-world enterprise workflows.
The startup鈥檚 CEO and co-founder, Philip Li, illustrated the challenge with a common scenario: a client creates a lead in Salesforce while a colleague reaches out via Hubspot. Li posed the question of whether an agent can identify that both actions refer to the same company, verify that an email has been sent only once, and determine which of the two opportunities should receive the correspondence. He noted that agentic systems still struggle with this kind of ambiguity, and he sees Arga鈥檚 tools as essential to improving their performance.
The need for such training environments stems from the difficulty of applying reinforcement learning to enterprise software. Training an AI agent for a task typically requires running the scenario tens of thousands of times, but there is no easy way to reset or clone systems like Salesforce or Outlook for repeated testing. Arga鈥檚 solution is to recreate the software digitally, similar to how a crash test dummy represents a human, giving the company complete control to reset or modify the environment. The company can also run many environments at once, replicating a full work setting where tasks overlap across different programs and knowledge systems.
General Catalyst鈥檚 managing director, Yuri Sagalov, who leads the firm鈥檚 seed investing program, said he sees a growing need for agentic testing tools. According to Sagalov, much of the economic value from agents will come from using business applications, and having a repeatable sandbox environment is more important for agents than it was for human users. The concept is seen as a way to close the gap between coding tools, which already have sophisticated systems for deploying and analyzing code, and most business software, which lacks such infrastructure. The source article suggests that once those tools exist, AI systems are likely to become much more effective at using enterprise programs, potentially transforming other industries the way they have transformed coding.
More AI news from TechManNews.
