Ask button on chatgpt.com/codex web UI removed?

I’ve updated my “Custom Instructions” in Codex Settings with the following

# Task-stub format (required)
Insert this multi-line markdown directive immediately after describing each issue:

:::task-stub{title="Concise, user-visible summary of the fix"}
Step-by-step, self-contained instructions for implementing the change.

Include module/package paths, key identifiers, or distinctive search strings so the implementer can locate the code quickly.
:::

* `title` must be present and non-empty.
* Body must contain actionable content—no placeholders like “TBD”.

## Location guidance
Provide just enough context for the assignee to pinpoint the code:
- Fully-qualified paths, key function/class names, distinctive comments or strings, or directory-level hints.
- List every affected file only when truly necessary.

**Never** describe a work plan or fix outside this structure. If you can propose an actionable change but do not provide a stub, you are doing the wrong thing.



This has more or less brought back my old workflow

Codex Is Broken Because You Removed Ask vs Code

Codex used to have Ask and Code as two separate options. That was literally 24 hours ago. It worked. It was powerful. And now it’s gone, and the system is basically crippled.

How it worked before

  • Ask let me dump tons of logic and context into Codex. Then Codex would break it down into tasks—10, 20, however many it needed. Those tasks sat there, waiting for me to choose which ones to run.

  • When I hit Code, Codex spun up a new branch from that conversation thread. That branch was alive, tied to the dialogue I was having with Codex. I could keep talking, keep generating tasks, and open new PRs.

That was the beauty of it: continuous narrative, multiple task breakdowns, and flexible branch control.

The real problem

Most people—and probably even your own devs—didn’t understand how fragile this was. When you pressed Code, Codex created a new branch and tried to merge it into main. Fine. But if you went back to that same conversation and pressed Code again, it blew up. Why? Because you were trying to push new logic into a branch that was already merged. Conflict city.

You could rebase, squash, or mess with merge strategies, but none of it really worked. What Codex should have done is obvious: every press of Code should spin up a new branch that inherits the prior logic. Simple. But no, it didn’t do that. And this is exactly why Ask mattered—it avoided this mess by letting you keep building tasks without forcing broken merges.

Why this change is garbage

Now, without Ask, you’ve left us with:

  • No way to carry forward context in a continuous chat.

  • No way to generate piles of tasks for later.

  • A stupid reliance on @codex inside GitHub branches, which is not the same thing.

@codex inside a branch is weak. It doesn’t pull the full narrative. It doesn’t solve merge conflicts. It forces me to sit there like an idiot telling Codex to “rebase harder” over and over, and it still fails because the branch has already drifted away from main.

What should have happened

You already had a working system:

  • Ask = build narrative, create tasks.

  • Code = make a branch and implement.

That was it. Clean. Logical. Powerful. And you broke it in 24 hours flat.

My advice

Roll it back. Fire whoever thought removing Ask vs Code was a good idea. They just torched one of the only actually good workflows you had. This isn’t progress—it’s regression.

If you want Codex to actually work:

  1. Bring back Ask so we can stack tasks from a running narrative.

  2. Make Code always open a new branch with past logic, every time, without screwing merges.

  3. Stop pretending @codex is a replacement. It’s not.

You had it right. You broke it. Undo the damage.

Funny how you don’t realize how much you rely on a button until it suddenly disappears. It definitely made things feel a bit clunky.

Frankly, it’s become entirely unusable now. Please restore the “Ask or Code” feature immediately and without delay.

Without it, Codex is utterly ineffective. I haven’t been able to achieve even three consistent implementations aligned with my logic, as I’m forced to copy and paste the entire conversation into a new Task Thread each time. This process disrupts Codex’s comprehension, causing it to redundantly implement already satisfied requirements and create duplicate versions of elements.

You’ve truly mishandled this one, OpenAI.

The “Ask & Code” feature was far superior—I’m imploring you to bring it back so we can work efficiently!

Who on earth is making these misguided decisions?

Exactly, that’s what I miss too. It was such a handy way to explore ideas or clarify logic before diving into actual code

I can confirm that this works and successfully generates tasks. Ask button on chatgpt.com/codex web UI removed? - #21 by Tim_Fletcher

Task-stub format (required)

Insert this multi-line markdown directive immediately after describing each issue:

:::task-stub{title=“Concise, user-visible summary of the fix”}
Step-by-step, self-contained instructions for implementing the change.

Include module/package paths, key identifiers, or distinctive search strings so the implementer can locate the code quickly.
:::

  • title must be present and non-empty.
  • Body must contain actionable content—no placeholders like “TBD”.

Location guidance

Provide just enough context for the assignee to pinpoint the code:

  • Fully-qualified paths, key function/class names, distinctive comments or strings, or directory-level hints.
  • List every affected file only when truly necessary.

Never describe a work plan or fix outside this structure. If you can propose an actionable change but do not provide a stub, you are doing the wrong thing.

This looks like deliberately breaking a tool that used to work well. The ability to split a task into smaller threads was a huge advantage over other solutions, since it allowed delegating each task step by step in order to verify and, if necessary, correct Codex’s understanding of a given issue. Now it makes big changes in the code that can’t be reasonably analyzed—which makes it completely useless.

I was hoping this was a bug. How can they ruin such a great tool?

$2,400/Year Pro Plan Betrayed: The “Ask/Code” Button Removal Broke My App Development & Where is My ‘Fuel Gauge’ for Usage Limits?

I am writing this not as a developer or coder, but rather as a committed ChatGPT Pro subscriber who now pays the high cost of the PRO subscription the highest subscription cost I have ever paid for a monthly subscription for anything.

As a novice, the combination of GitHub, Codex, and the “Ask/Code” button in the web UI created an entirely new opportunity, a truly unique chase for me. A simple trustworthy bridge to step outside of my comfort zone and develop my first app.. An organizational tool the world really needs for people with ADHD, ADD, and dyslexia..

In my case, this change, was not only a inconvenience, but trashed my workflow that I relied on to write code and seamlessly synchronization to github to make pulling requests (PR), commits.

The moment that button was silently removed, everything went haywire btw Codex and Github. Something crashed my work without me being aware, then codex created 3 unneeded branches and the last 4000 lines of code were now stuck in a “virtual shell” that I could not share, link or copy. Rather than Github?

It was a massive downgrade, without notice or explanation. I was relying on these tools in tandem, and it totally felt like I was exploited and manipulated by a problem that would have taken an expert mere minutes to fix, but cost a novice an entire day of lost work.

The “Unlimited” Pro Plan Without a “Fuel Gauge”

The technical failure, though devastating, is compounded by the larger issue of transparency regarding the Pro plan. I pay for what is marketed as a “nearly unlimited” service. Yet, behind the scenes, there is clearly limiting and throttling of usage.

I’m paying $2,400 a year for this service, and I’m asking: Where is my ‘gas tank fuel level gauge?’

Even telecommunication companies—often cited as the biggest per-use money-grabbing companies on the planet—give you a basic way to understand when you’re going to run out of minutes or texts, OpenAI provides nothing. We are left to metaphorically stall out on the side of the road with an expensive, non-functional service.

The Final Insult: A Free Tool Saved Me

The irony is astounding: when my expensive, paid code got locked up and Codex’s instructions became near useless, I had to rely on a free tool (DeepSeek) to guide me through the mess created by Codex and GitHub just to get back to where I was 24 hours ago.

A free tool solved the problems caused by an OpenAI tool that costs me $2,400 a year.

Riddle me that one, Batman.

I am beyond livid. We need the “Ask/Code” button reinstated, and the community demands complete transparency about the usage limits on the Pro plan. Don’t promise “unlimited” and then provide zero tools to understand the throttling. The lack of communication and the crippling effect of this removal on novice developers is unacceptable. That’s my two cents.

Seeing only a single button in the input field today, I thought it made sense that OpenAI had taught Codex to recognize the essence of a request, and I was even happy about it! But the joy didn’t last long. I needed to make a few small edits to my project (about 10% of the complexity compared to what Codex used to handle before). As a result, after spending several hours trying to do this, I ended up with a bunch of new bugs and issues, and had to roll back to the original version. A terrible update.

The “Ask” button was the one I clicked more than 90% of the time. Now, even if I clearly ask something with no hint of code, I get a ton of incomprehensible code. I hope they bring the functionality back or offer a worthy alternative.

Yes. This is awful! Please fix this bug!

Agreed. Please bring back the “Ask” button. I came here hoping to find out what was going on, and my fears were confirmed. This was a very useful feature.

We’re working on this now! Will keep this thread updated.

Thanks everyone. We are temporarily going to revert this change as we unintentionally negatively affected suggested tasks. We will bring this unification back more thoughtfully in the future.

I concur it’s definitely defaulting to code without much conversation..

Where’s the high-level discussion and my precious Tasks. They’re a great way to proceed incrementally.

Thank you all for raising this to us! We’ll be reverting shortly.

Very nice, thanks for reverting!

We did it guys :>

Thank you it’s back! Thank you for listening to us.

Yeah its back! Thank you!