Arga Labs has raised a $10 million seed round led by General Catalyst, with participation from Box Group, Emergence, Gradient, and SV Angel, to build training environments that help AI agents operate reliably within enterprise software such as Salesforce, Workday, and email systems.
Unlike typical testing setups that rely on stateless API endpoints, Arga Labs constructs full digital twins of enterprise programs, replicating permission systems and webhooks to more accurately train agents across interconnected platforms. CEO and co-founder Phillip Li said the goal is to help agents handle ambiguous, cross-system scenarios, such as recognizing when a lead created in Salesforce and a separate outreach through HubSpot refer to the same company, tasks that remain difficult for current agentic systems.
Traditional reinforcement learning approaches require running scenarios repeatedly, which is largely impractical with real enterprise software due to the difficulty of resetting systems like Salesforce or Outlook. Arga’s simulated environments solve this by allowing resets and parallel training at scale, mirroring how coding tools enabled faster AI progress through easy testing and iteration, a capability largely absent for business applications until now.
Yuri Sagalov, managing director at General Catalyst and head of its seed program, said much of the economic value from AI agents will come from their use within business applications, making repeatable sandbox environments increasingly essential as agents take on more complex, real-world tasks.
Featured image: Credit: Arga Labs
Arga Labs has raised a $10 million seed round led by General Catalyst, with participation from Box Group, Emergence, Gradient, and SV Angel, to build training environments that help AI agents operate reliably within enterprise software such as Salesforce, Workday, and email systems.
Unlike typical testing setups that rely on stateless API endpoints, Arga Labs constructs full digital twins of enterprise programs, replicating permission systems and webhooks to more accurately train agents across interconnected platforms. CEO and co-founder Phillip Li said the goal is to help agents handle ambiguous, cross-system scenarios, such as recognizing when a lead created in Salesforce and a separate outreach through HubSpot refer to the same company, tasks that remain difficult for current agentic systems.
Traditional reinforcement learning approaches require running scenarios repeatedly, which is largely impractical with real enterprise software due to the difficulty of resetting systems like Salesforce or Outlook. Arga’s simulated environments solve this by allowing resets and parallel training at scale, mirroring how coding tools enabled faster AI progress through easy testing and iteration, a capability largely absent for business applications until now.
Yuri Sagalov, managing director at General Catalyst and head of its seed program, said much of the economic value from AI agents will come from their use within business applications, making repeatable sandbox environments increasingly essential as agents take on more complex, real-world tasks.
Featured image: Credit: Arga Labs
