Improve Support for Long-Horizon, Multi-Session Development Tasks

[EricGT]

What variant of Codex are you using?

VSCode Extension

What feature would you like to see?

Codex works well for short, focused tasks, and Plan Mode helps structure near-term work. However, many real-world efforts unfold over extended periods and multiple sessions.

Examples of longer-horizon workflows include:

• Software & Engineering

  • Large multi-file refactors completed in stages
  • Staged optimization (e.g., instrumentation → data collection → rebuild)
  • Developing parsers against real-world input where the data may deviate from formal specifications and must be refined iteratively

• Research & Science

  • Designing and refining experiments over multiple hypothesis cycles
  • Analyzing evolving datasets with iterative model adjustments

• Writing & Knowledge Work

  • Writing a fiction book across chapters and revisions

The OpenAI Cookbook example on long-horizon tasks reflects this style of work, where progress unfolds over many coordinated steps rather than a single interaction.

It would be valuable for Codex to better support the creation and execution of long-horizon tasks, which often appear straightforward in concept but prove significantly more complex in practice.

EDIT

The original long-horizon tasks link is now dead:
https://github.com/openai/openai-cookbook/blob/main/examples/codex/long_horizon_tasks.md
per openai/openai-cookbook@2de2dbd
https://developers.openai.com/blog/run-long-horizon-tasks-with-codex

Additional information

No response

[etraut-openai]

Thanks for the clear problem statement. I'm curious whether you have any solutions in mind.

[EricGT]

Thanks for the reply.

I'm curious whether you have any solutions in mind.

Not at the moment.
Will keep an eye open and update if I find something.

ChatGPT Pulse comes in handy for keeping updated on spotting new possiblities.

[EricGT]

Thanks for the clear problem statement.

Thanks for the direction.

We also recommend that feature requests focus on the problem statement rather than a solution. Describe the use case and what problem you'd like to see solved. There are often multiple ways to address a given problem.

[etraut-openai]

This feature request hasn't received enough upvotes, so I'm going to close it.

[EricGT]

@etraut-openai

Thanks for the heads up on closing, I just hope more people see the benefit of Long Horizon task.

Do you know of any other places where others are discussing Long Horizon tasks?

[etraut-openai]

@EricGT, I'm personally working on some new features and primitives to help with long horizon tasks. It's an important area of development.

There are many dimensions to this problem. I suspect that the reason this feature request didn't get any upvotes is because it was not specific enough. Are there some specific capabilities that you think we should prioritize? If so, look for existing feature request and post to those threads or open new ones. That's the best way to provide input and help us prioritize this work.