OpenAI Agents API Artifact Contract: Make Long-Running Agent Work Reviewable

Last Updated on September 25, 2026 by Editorial Team Author(s): Ethan Mark Originally published on Towards AI. OpenAI Agents API Artifact Contract: Make Long-Running Agent Work Reviewable OpenAI Agents API Artifact Contract An agent that says “done” has not necessarily delivered anything a person, service, or next agent can safely use. An artifact contract turns a cloud-agent run into a package with a known shape, proof, and next step. Treat the output boundary as part of the product, not an afterthought. A long-running agent can investigate an incident, write code, analyze a dataset, and produce a polished final message. Then a reviewer asks a basic question: where is the actual work? The answer is often a hunt through a temporary workspace, a chat transcript, or a folder full of files with unclear names. That gap becomes expensive as soon as agents work asynchronously or hand work to other people. A manager cannot approve a mitigation that has no evidence bundle. A service cannot reliably ingest a report whose filename changes on every run. A second agent cannot tell whether an empty folder means “nothing found,” “job failed,” or “the first agent saved it somewhere else.” The ti…

≈