Is Codex falling behind Claude Code Desktop?

Codex
1

I have the feed is that OpenAI is lagging more and more behind Claude Code Desktop. Things like IOS simulator right in the app it very nice. But many other simple things like /reset to clear the screen and context. Just look at their weekly updates: What's new - Claude Code Docs

Codex team should release at least some kind of roadmap with things which are planned and start listening to people in this community. Feature requests are ignored - as in nothing specific is planned based on the requests.

And people in the community must get more active if they want something to change. It’s very quite in here.

2

I agree with the general point. The feature gap matters less than the lack of visibility into what the Codex team is prioritizing. Even a simple roadmap with upcoming UX improvements, simulator support, context controls, and other frequently requested features would make it easier for users to know where things are heading.

3

I use ClaudeCode as either a project manager or a reviewer. Best use of it’s minimal tokens per week. Sometimes, I let a project GPT act as project manager, have Codex code, then Claude review code. Sometimes, I have them working on projects at the same time, but I have a very specific agent kit I set up for that.

4

@methuselas Would you mind sharing more about your workflow with multiple agents and the reasoning?

For my solo work I use only Codex right now (I tried using Claude and Codex, but was not very succesful).

I usually do a general task creation with broad idea/context and store it as a file. Then I clarify the task by letting the agent interview me to find to think of (all possible) issues/blockers/conflicts/etc which might come up to create good specification for this task and then update the task file with all the details.

Then I let another agent read the file, do one more session of questions before starting the task. I work in 4-6 worktrees which run 4-6 of these tasks in parallel. Each task can be tested as a whole (in 95% of the cases), so after each task is done, it’s put in review mode and then it’s reviewed, it’s then archived with all context before the task was started and after it was finished stores in this task. I of course let the agent review own work and have a definition of what is expected. After a while a review each result manually as well.

PS: I use Backlog.md for kanban style task management (all in repo as markdown with a little tooling and UI around it). But before I used just markdown files with folders for task states.

5

You can keep up with codex’s new features here:

As for the new features, the best place is the issues section on github, since codex cli and its internal core (app-server) are opensource.

There you can see what are the most requested features, vote on an existing one (thumbs up) or suggest new features and report issues:

6

I’ll do you one better and just give you my agent kit. :grin:

The readme.md explains it, but essentially you use it to coordinate tasks in the same project, so multiple Agents can work in the same repo, committing and pushing their own work, but keep it organized so you know who did what, when, where it worked, what failed, etc. Also comes with a very rudimentary version of PASS (Pattern Analysis Skill System), so it can also save drills and patterns it discovers, while coding. My Dungeon Forge project is being built by me, Claude, Codex, a local Qwen 3.5b and using a GPT project as the project lead.

What you want to do is work with an Agent whose sole purpose is to work with you on arranging your project, be it vibe code, project notes, GDDs, etc. That Agent will then help you split up assignments in a worklog. Then, you have the same Agent issue jobs to specific Agents of your choosing (or let the Agent, itself do it. Claude is outstanding in this, but I’d rather use him for code) in a copy pasta. Once you get the copy pasta, you hand it off to the Agent it was assigned to and let it’s do it’s thing. Agents won’t touch other assignments, if they’re assigned to another Agent, unless you override it.

When each assignment is finished, it will create a folder based off the assignment using the “domains” template. These get dropped in the docs folder, with the worklogs. The folder will have separate markdowns for decisions, failures, next steps, a worklog and an index. It then closes the assignment and will commit it (and push if you allow it). Finally, it closes out the assignment in the main worklog and waits for new [OPORD].

You can also use GPT projects, if you want to maximize your output. Just copy a repo archive, sans the .git folder to it’s source files and hand it it’s own assignment. The downside to this is you’ll have to manually commit and push this one and the Agents who have access to your local repo, won’t see the updates, until you manually merge it. I mainly use this when I run out of tokens for Claude and Codex, but need work done.

Hope it helps. o7

-M

[EDIT] - Oh, one more cool thing is any Agent can update the main worklog.md if you let them. If, while working on something, it finds that it requires another assignment done, that’s not in the worklog, it can update the worklog.md, with what it needs and even assign it to another Agent. It’s better to use a Project Lead for this, though.

Agent Kit