Project-level orchestration: make Tasks an implementation detail
Nobody has claimed this yet.
- Dominant language
- Rust
- Stars
- 125k
- Forks
- 19.4k
- PR merge metrics
- PR metrics pending
Description
What variant of Codex are you using?
App
What feature would you like to see?
I would like Codex to have an optional persistent Project-level orchestration mode above individual Tasks.
This should not replace the current task-centric workflow. Advanced users who prefer to manage tasks, agents, models and worktrees directly should be able to continue doing so.
The Project mode would be particularly useful for less experienced developers, domain experts, and vibe-coding users who know what they want to build but do not necessarily know how to orchestrate coding agents efficiently.
In this mode, the user would discuss the goal, constraints and architecture with Codex in a persistent Project conversation, approve the plan, and then let the Project-level agent manage execution: create and maintain issues, decide sensible task boundaries, choose appropriate models and reasoning levels, coordinate dependencies, manage worktrees, request independent reviews, and update the plan when circumstances change.
Tasks and agents would still exist, but they would become implementation details managed by Codex rather than something the user must operate manually.
A useful analogy is a contractor: I discuss the project, priorities and changes with the contractor; I am not expected to coordinate every individual worker myself.
In short:
User manages project intent → Codex manages the agent workflow → task agents execute the work.
Additional information
I think this becomes increasingly important as the models improve. In my experience, model capability is no longer the main entry barrier; learning how to use the models efficiently is.
New users can spend significant time and usage learning task decomposition, context management, model routing, Git/worktree workflows and review strategies. Poor orchestration can consume a large amount of tokens without improving the result.
Plan mode already provides the beginning of the right interaction model: discuss a plan, approve it, then execute. I think the natural extension is:
Project conversation → Plan → Approval → Orchestration → Execution → Review → Plan update
The Project agent should optimise for quality, reliability, cost and latency, rather than maximising the number of agents or parallel tasks.
Related requests already address parts of this problem:
- #33186 — bounded task orchestration
- #36790 — model/reasoning routing for created tasks
- #14923 — cross-thread orchestration
This proposal is intentionally one level above those features. It is not mainly about exposing more orchestration controls to the user. It is about making Codex responsible for operating those controls on behalf of the user.
I see this primarily as an optional accessibility and efficiency layer, not a replacement for expert workflows. It could allow users who currently avoid multi-agent workflows, model routing, worktrees or structured reviews because they are difficult to manage to benefit from those capabilities indirectly through Project mode.
This could be especially valuable for vibe coding. The models and tools are already powerful, but users still need considerable knowledge to combine them efficiently. A Project-level contractor could lower that entry barrier while preserving full manual control for advanced users.
The primary interface for agentic coding does not always need to be the Task. For users who prefer it, Codex could offer the Project as a higher-level interface, while keeping Tasks available underneath for direct control.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
This is a high-level proposal for an optional Project-level orchestration mode; it names no files, tests, or implementation entry points. Start by reading related issues #33186, #36790, and #14923, then clarify the product scope and acceptance criteria before implementation. Done would require an agreed design covering planning, task and agent management, reviews, and plan updates.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- rust
- Domain
- ai, devtools
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Active
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100