Arga is building a better way to train enterprise AI agents
techcrunch ·
Arga has raised $10 million in seed funding led by General Catalyst, with participation from Box Group, Emergence, Gradient, and SV Angel. The startup builds training environments for enterprise software such as Salesforce, Workday, and email clients, creating full-scale digital twins of these programs with permission systems and web hooks intact. This approach allows AI agents to be trained across multiple systems in a more robust way than standard stateless API testing environments. CEO Philip Li gave an example of an agent needing to determine whether leads in Salesforce and Hubspot refer to the same company and manage email outreach correctly. Training agents through reinforcement learning is difficult for enterprise software because systems like Salesforce cannot easily be reset or cloned for repeated scenarios. Arga's digital recreations are easy to reset, modify, and run in parallel, enabling complex multi-system training. This resembles the tools available for coding that have accelerated AI coding progress, but such tools are largely missing for business software. General Catalyst's Yuri Sagalov said repeatable sandbox environments are increasingly important for agents using business applications.
techcrunch