Mercor Buys Deeptune to Own the 'Environments' Where AI Agents Train

The race to train AI agents is turning into a race to own the rooms where they practice. On July 9, 2026, Mercor — an AI data and evaluation company running at roughly $2 billion in annualized revenue and last valued at $10 billion — said it is acquiring Deeptune, a startup that builds simulated software environments for agents to learn in. Terms were not disclosed; Deeptune's team of about 20 joins Mercor in New York.

What Deeptune makes

Deeptune builds what founder Tim Lupo calls "flight simulators for AI doing work" — high-fidelity, interactive replicas of real enterprise applications like spreadsheets, Salesforce and Slack. Inside them, an AI agent takes actions across a multi-step task, gets graded against a reward signal, and improves through reinforcement learning, rather than absorbing a static dataset. It is a sandbox where an agent can fail safely before it touches a company's live systems.

Why environments matter

Mercor's own business is the other half of that equation: it pays a network of vetted domain experts to author the tasks and scoring rubrics that judge whether a model did a job correctly, work it sells to labs including OpenAI and Anthropic. As Mercor frames it, it builds the tasks and rubrics; Deeptune builds the apps those tasks run inside. Together they cover more of the stack for training agents that actually operate software.

The insider angle

The deal has an unusual backstory. Deeptune raised a $43 million Series A led by Andreessen Horowitz on March 19, 2026, and among the angels was Mercor co-founder and CEO Brendan Foody. Buying the company, Foody told Fortune, "was in a lot of ways the main motivation, actually" for writing that check about three months earlier — a candid admission that doubles as a governance talking point given the undisclosed price.

The land grab

Environments have become one of AI's hottest categories, with dozens of startups chasing the title of "Scale AI for environments" and labs reportedly weighing nine-figure budgets for them. a16z partners investing in the space have argued that "if the last decade of AI progress was driven by better datasets, the next decade will be mostly driven by better environments." Mercor, itself only founded in 2023, is now consolidating that thesis by acquisition.