Arga Labs has raised a $10 million seed round to create realistic training grounds for AI agents that operate business software. General Catalyst led the funding, with Box Group, Emergence, Gradient and SV Angel participating.
The startup builds digital twins of products such as Salesforce, Workday and email clients. Unlike a simple mock API, each replica preserves elements including permissions and webhooks. Teams can reset the environment, alter it and run many copies at once, making repeated reinforcement-learning trials practical without disturbing a real corporate system.
That matters when a task crosses several tools. An agent may need to recognize that a Salesforce lead and a HubSpot contact represent the same company, choose the correct recipient and avoid sending duplicate email. These ambiguities are difficult to test at scale when the underlying applications cannot easily be returned to a clean starting state.
Coding agents benefited from mature systems for testing, versioning and undoing changes. Arga is trying to supply an equivalent layer for enterprise applications. Its promise depends on how faithfully the replicas reproduce the permissions, state changes and edge cases agents will encounter in production.