These demo agents can survive process crashes, scale to millions of executions in parallel and have easy-to-implement human interactivity. Just add a couple of Python decorators to your OpenAI agent code, run Temporal workers and you're ready to go.
Check the video I did with OpenAI showing this in action (it's linked in the repo).
OpenAI actually use us for ChatGPT Images and also their Codex code writing agent so I figure this may be a viable path for others to code something easy that's also reliable at scale.
Happy to answer questions.