OpenAI wants its agents to handle your contracts, not just your chat window.
OpenAI and Ironclad, a contract lifecycle management platform used by legal and procurement teams, are jointly training and evaluating AI agents on complex contracting workflows. The work is framed as a way to advance "computer use" - agents that operate software directly rather than just generating text - for professional tasks. Contracting is a deliberately hard testbed: it requires reading dense documents, checking terms against policy, and acting in the right sequence inside real software, not just answering a question in a chat box. Both companies are treating this as an evaluation exercise, measuring how well current models perform before anyone hands an agent a live negotiation.
This matters because it signals where OpenAI expects the next round of agent competition to happen: inside back-office software, not inside search bars. Contracting is full of compliance stakes and edge cases, so a credible result here is a tougher proof point than another coding or browsing demo. It also gives Ironclad a reason to look like the place legal teams go when they want to experiment with AI, rather than build it themselves.
OpenAI already ships a computer-use agent called Operator, and the industry has a long history of promising AI that handles paperwork and then quietly shipping autocomplete - whether this effort clears that bar will show up in signed contracts, not in a press release.