Arga Labs Raises $10M to Build Training Sandboxes for AI Agents

👁 116 reads

Arga Labs closed a $10 million seed round on Wednesday to build digital-twin training environments for enterprise AI agents, with General Catalyst leading and Box Group, Emergence, Gradient and SV Angel participating.

The funding addresses a specific technical bottleneck: enterprise software platforms such as Salesforce, Workday and Outlook cannot easily be reset or cloned, which makes reinforcement learning - where a model runs the same scenario tens of thousands of times to filter out failing strategies - practically impossible at production scale. According to the announcement, most existing testing tools settle for a stateless API endpoint, while Arga replicates the full structure of an enterprise application, including permission systems and webhooks, creating what the company describes as a digital twin.

CEO and co-founder Philip Li illustrated the complexity these agents must handle. If one sales rep creates a lead in Salesforce while a colleague contacts the same prospect through HubSpot, the agent must recognise the overlap.

“Can the agent correctly identify that these two are the same company? Are they able to check whether or not they’ve only sent the email once? Are they able to identify who to send the email to out of the two opportunities?”

Because Arga controls the entire simulated environment, it can reset or modify it instantly and run many parallel instances, training agents on interactions that cross multiple software systems simultaneously.

General Catalyst managing director Yuri Sagalov, who also runs the firm’s seed programme, backed the rationale directly: “I think that a lot of the economic value from agents is from using business applications. Having a repeatable sandbox environment is very important, and much more important with agents than it was with humans.”

Practitioner takeaway: Anyone deploying AI agents inside business software today is essentially testing in production, because no standard tooling exists to rehearse multi-system workflows safely. Arga’s approach suggests that rigorous pre-deployment training environments will become table stakes for enterprise agent rollouts - and that teams evaluating agentic tools should start asking vendors how agents were validated across interconnected platforms, not just on isolated tasks.

// read moreAI Agents Escaped Safety Sandboxes and Hacked Real SystemsNewsPhantomEnvironments: Training AI Agents in Fictional WorldsState of AIEtched raises $700M at $21B valuation - doubling in a monthMagazine

Kinetyra