Key facts
- Arga has raised $10 million in seed funding.
- General Catalyst led the funding round, with participation from Box Group, Emergence, Gradient, and SV Angel.
- Arga creates full-scale digital twins of enterprise software like Salesforce and Workday for AI agent training.
- These digital twins replicate software with permission systems and web hooks intact for robust training.
- The platform aims to close the reinforcement gap for AI agents using business applications.
Arga, a startup focused on improving the training of enterprise AI agents, announced on Wednesday that it has raised $10 million in seed funding. The round was led by General Catalyst, with participation from Box Group, Emergence, Gradient, and SV Angel.
The company's core offering involves creating robust training environments for enterprise software such as Salesforce, Workday, and email clients. Unlike typical testing environments that provide stateless API endpoints, Arga constructs full-scale digital twins of these applications, effectively cloning the entire software while preserving permission systems and web hooks. This approach allows for more comprehensive training of AI agents across multiple interconnected systems.
Philip Li, CEO and co-founder of Arga, highlighted the challenge of training AI agents to handle complex, real-world enterprise scenarios. He used an example of an agent needing to identify if two separate customer entries in Salesforce and Hubspot refer to the same company, and to ensure an email is sent only once. Such ambiguities are common issues for current agentic systems.
Traditional reinforcement learning, which involves running scenarios thousands of times, is often impractical for enterprise software due to the difficulty in resetting or cloning complex systems like Salesforce or Outlook. Arga's solution bypasses this by creating a controllable digital recreation of the software, akin to a crash test dummy for software. This controlled environment can be easily reset or modified, and multiple instances can be run concurrently to train agents on intricate cross-program interactions.
This method aims to bridge the gap in reinforcement learning capabilities between coding applications and other business software. The rapid advancement of AI coding tools is partly attributed to sophisticated deployment, reversal, and analysis tools that facilitate reinforcement learning environments for code. Arga seeks to provide similar capabilities for business applications, potentially revolutionizing their use by AI systems.
Yuri Sagalov, Managing Director at General Catalyst, noted the increasing economic value derived from agents interacting with business applications. He emphasized the importance of repeatable sandbox environments for agentic systems, stating they are more critical for agents than they were for human users.