The companies are training and evaluating agents, but have not described tasks, scores, or customer availability.
OpenAI and Ironclad are working on AI agents for complex contracting workflows, OpenAI said October 6. The companies present the effort as a way to advance computer use for professional work.
An AI agent is software that can pursue a task through several steps, rather than only answering one question. In this case, the proposed setting is contract work, where useful automation could involve moving through a series of connected actions.
The announcement describes both training and evaluation. Training is the process of improving an agent’s behavior. Evaluation means testing whether it performs the intended work reliably. Together, those steps could make contract workflows a focused test for AI systems that operate professional software.
The practical question is not simply whether an agent can read contract language. It is whether the system can handle a full workflow accurately, recognize when it is uncertain, and avoid creating costly mistakes. The announcement does not say which contracting tasks the agents will perform or how much human review they will need.
It also does not provide evaluation metrics, results, error rates, training-data details, or information about operational availability. OpenAI and Ironclad have not established whether this is a research effort, a pilot, or a system customers can use.
For engineers and product leaders, the next useful milestone is a task-level evaluation: clearly defined contract jobs, success measures, failure cases, and the human checks required before an agent’s work is accepted.