Making AI agents work is turning out to be a lot harder than many companies expected. Luckily for them, there\u2019s help on the way: A new crop of startups is finding better ways to test and train agents before they are deployed, particularly on the complexities of the modern enterprise.<\/p>\n
Arga<\/a> is one such company. The startup on Wednesday said it had raised $10 million in a seed funding round that was led by General Catalyst, with participation from Box Group, Emergence, Gradient and SV Angel.<\/p>\n Arga builds training environments for enterprise software like Salesforce, Workday, and email clients. Where most testing environments offer a stateless API endpoint, Arga builds a full-scale digital twin of the program, effectively cloning the entire software with permission systems and web hooks intact. The result is a more robust way to train agents across multiple systems.<\/p>\n The startup\u2019s CEO and co-founder Philip Li illustrated his product with the example of a prospective client creating a lead in Salesforce while a colleague reaches out separately through Hubspot.\u00a0<\/p>\n \u201cCan the agent correctly identify that these two are the same company?\u201d Li says, \u201cAre they able to check whether or not they\u2019ve only sent the email once? Are they able to identify who to send the email to out of the two opportunities?\u201d<\/p>\n Agentic systems still struggle with this kind of ambiguity, and Li sees Arga\u2019s tools as critical to helping them improve.<\/p>\n Normally, an AI agent could be trained for a task like this through reinforcement learning (RL): running the scenario tens of thousands of times and letting only the successful strategies through. But the nature of enterprise software makes that scale of testing nearly impossible. There\u2019s no easy way to \u201creset\u201d a system like Salesforce or Outlook when you need to run the same scenario again, much less clone it.\u00a0<\/p>\n Arga\u2019s solution is to digitally recreate that software by replicating its structure the way a crash test dummy is supposed to stand in for a person. Because Arga has complete control over the environment, the resulting recreation is simple to reset or modify. The company can also run many environments at once to train agents on the complex interactions between different programs. The idea is to replicate a person\u2019s full work environment, with specific tasks overlapping across different programs and knowledge systems.<\/p>\n