Mercor raised $750 million at a $20 billion valuation, the company announced Nov. 10. It says the money will expand the work it does to help companies evaluate and train AI agents on their own tasks. Investors in the Series D include Benchmark, NVIDIA, Salesforce and Zoom. Mercor says the round included an employee tender. Mercor’s funding announcement and the company’s account of the round
What does an enterprise get from a test built for its own work? Take invoice handling. An agent can recognize a vendor and an amount in a public test. The business might need it to flag an invoice paid twice, route a mismatch to the right person and leave a record the next shift can follow. Mercor says it builds evaluations with domain experts around those kinds of company-specific standards, then uses the tests to guide additional training. The company says its earlier acquisition of Deeptune adds practice environments modeled on enterprise software. Mercor’s description of the service
For a buyer, the first deliverable worth inspecting is the test. Choose a workflow with a clear cost when it goes wrong. Ask the people who do the work to bring ordinary cases and the exceptions that consume their time. Put some cases aside to see whether the agent improves on examples it has yet to see. If the company can keep those tests, it can run the same comparison when the model or the cost of using it changes.