Why does copilot stop in the middle of the work?

[Rand-Random]

Select Topic Area

Question

Copilot Feature Area

General

Body

As you can see in the screenshot below (from Visual Studios Copilot integration)

it just stops and asks me if it was helpful and again and again I have to trigger to continue to do its job by asking "done?" or similar.

[Sharan-Kumar-R]

Copilot Stopping Mid-Work in Visual Studio

Hi @Rand-Random, thanks for reporting this!

From your description and screenshot, it appears that GitHub Copilot’s inline chat in Visual Studio is pausing to ask for confirmation (“Was this helpful?”) before continuing.

This usually happens when:

• Copilot Chat is set to require explicit confirmation between steps. When you ask for a large-scale change across multiple files, Copilot breaks it down into smaller chunks.
• The extension’s conversation flow times out or loses context.
• An older version of the Copilot extension is in use.

---

Possible Fixes

1. Update Visual Studio and Copilot extension
   Make sure you are on the latest versions of both.

2. Switch to Chat View
   Instead of inline mode, use the Copilot Chat panel so it can continue responding without prompting after each step.

3. Check Copilot Labs Settings
   If “Inline Feedback Prompts” is enabled, try disabling it.

4. Sign Out and Sign In Again
   This can reset any session-related issues.

---

Additional Information Needed

If the problem persists, please share:

• Visual Studio version
• GitHub Copilot extension version
• Whether this happens in all files or specific languages/projects

This will help determine if it’s a bug or an intended interaction flow that needs improvement.

[Rand-Random]

it appears that GitHub Copilot’s inline chat in Visual Studio

not the inline chat, but the sperate chat window (agent mode)

Copilot Chat is set to require explicit confirmation between steps.

a confirmation of some sort is comming up "was this helpful?", but this should NEVER be asked in the middle of the work, only at the very end it should be asked, not matter how long or complex the current task is

The extension’s conversation flow times out or loses context.

well that sounds like a bug, and should not happen

An older version of the Copilot extension is in use.

extension is built-in and no seperate extension is needed.
though, I noticed the old version were still installed but marked as incompatible with my visual studio version.

Update Visual Studio and Copilot extension
everything is up-to date.

Version 17.14.12
VisualStudio.17.Release/17.14.12+36408.4

extension is built-in, so no clue where I would find the version of it.

Switch to Chat View

already am, and you (or a bot) didn't correctly pick it up from the provided screenshot

Check Copilot Labs Settings

if that solves my problem, that it doesnt fix the underlying issue at hand
couldn't find the proposed setting, am I looking at the wrong place, all I can see is:

Sign Out and Sign In Again

since the problem is persistant, many restarts of visual studios and pcs - and active session should be long gone

Visual Studio version

Version 17.14.12
VisualStudio.17.Release/17.14.12+36408.4

GitHub Copilot extension version

extension is built-in, so no clue where I would find the version of it.

Whether this happens in all files or specific languages/projects

only work with c# projects and it happens everywhere

[rabbit-ninja]

@Rand-Random, This usually happens with large codebases, I ran into the same issue while using Copilot (as a paid user). I solved it by manually tagging the files instead of letting Copilot select them automatically. You can do this by including file names in the chat using # when prompting. This approach significantly reduces token usage and helps avoid the problem. Hopefully, future updates will handle large contexts more efficiently.

Alternatively, you could try using Gemini instead of OpenAI, since Gemini supports larger contexts. However, I believe the ability to switch models might only be available in their paid version (though I’m not entirely sure).

[NaeeemAbbas]

there may be multiple possibilities
1.Copilot uses the OpenAI API behind the scenes, and if your account hits its rate limit, suggestions may stop appearing or partially load.
2.If your internet connection is unstable or GitHub Copilot servers experience latency, it can cause incomplete suggestions.

Solution:

Check your network.

Restart VS Code or your IDE.

Try again after some time.
3.Copilot uses the current file + surrounding context to generate suggestions. If your file is very large or has many open tabs, it might struggle and cut off mid-suggestion.
4.If your GitHub token expired or your IDE lost connection to GitHub, Copilot might partially work or stop mid-way.
5.Other AI or autocomplete extensions (like Tabnine or IntelliCode) can conflict and cause interruptions.

[Rand-Random]

@rabbit-ninja
thanks for the input, I am using clauda sonnet 4, and yes I do have a paid subscribtion.

it was my guess that it is some sort of rate-limit mechanism, but IMHO if that were the case it should just sleep to decrease the amount of requests and continue with the task without user interaction, to be always on the edge and waiting for this feedback message is kinda annoying

one task which took ~30 minutes to complete (had to process many files), I was aked ~10 times

[rabbit-ninja]

that can be the reason i guess. large context creates this issue. Will have to dive into the copilot docs (if available)

[Rand-Random]

after only a few minutes of work (maybe 5)

here my prompt and the full context of the work

and it just stops, and I need to prompt "done?" and it will continue as if nothing happened
this happens again and again and again

edit:
while typing this message again same situation

@Sharan-Kumar-R just in case you didn't get my previous respond

[NaeeemAbbas]

there may be multiple possibilities
1.Copilot uses the OpenAI API behind the scenes, and if your account hits its rate limit, suggestions may stop appearing or partially load.
2.If your internet connection is unstable or GitHub Copilot servers experience latency, it can cause incomplete suggestions.

Solution:

Check your network.

Restart VS Code or your IDE.

Try again after some time.
3.Copilot uses the current file + surrounding context to generate suggestions. If your file is very large or has many open tabs, it might struggle and cut off mid-suggestion.
4.If your GitHub token expired or your IDE lost connection to GitHub, Copilot might partially work or stop mid-way.
5.Other AI or autocomplete extensions (like Tabnine or IntelliCode) can conflict and cause interruptions.

[Rand-Random]

@NaeeemAbbas
thanks for the input

you said "Copilot uses the OpenAI API behind the scenes" - I am using clauda sonnet 4 - how would I know which limit applies to me if any?
looking at the doc, of clauda sonnet 4
https://docs.anthropic.com/en/api/rate-limits

Our rate limits for the Messages API are measured in requests per minute (RPM), input tokens per minute (ITPM), and output tokens per minute (OTPM) for each model class. If you exceed any of the rate limits you will get a 429 error describing which rate limit was exceeded, along with a retry-after header indicating how long to wait.

I would hope that copilot respects the limits and reads the retry-after correctly and would automatically retry it.

Or at the very least give any form of feedback, currently there is no message, no exception, just nothing except for the frustrating part that it stops.

Try again after some time.

as I said a simple "done?" or "continue" is enough to trigger it to continue the task, and I dont even need to wait any amount of time, I can type it instantly after the stop and it will continue

internet connection

internet is stable on my side. full duplex business internet there should be no issue.

If your GitHub token expired or your IDE lost connection to GitHub, Copilot might partially work or stop mid-way

isnt there a logging anywhere to figure out the issue?

Other AI or autocomplete extensions (like Tabnine or IntelliCode) can conflict and cause interruptions.

Since I am not inline chat, but rather the chat window, I sure hope that no extension of mine is messing around with a random window.

[Pratap6130]

Copilot can sometimes stop mid-suggestion for a few reasons:

Context limit → It only looks at a certain number of lines/tokens, so long files or prompts may cause it to cut off.

Incomplete intent → If your comment or code isn’t specific enough, Copilot may generate part of an answer and stop.

Server or network hiccups → Occasionally, requests time out before the full response is delivered.

Editor integration → Some IDEs (VS Code, JetBrains) may show partial completions if Copilot decides the suggestion is “done.”

To reduce this: try breaking tasks into smaller steps, give clearer prompts (comments), and make sure your editor + Copilot extension are up to date. If it still happens often, you can report it directly to GitHub Support.

[pedro-salgueiro]

I think Visual Studio is missing this setting ( available in VS Code):

Agent: Max Requests
The maximum number of requests to allow Copilot to use per-turn in agent mode. When the limit is reached, Copilot will ask the user to confirm that it should continue.

If anyone can figure out if it can be set elsewhere please share, I've searched the registry and found nothing.

The behavior should be as long as you have tokens in your account it should finish the task that was given to it.

[Rand-Random]

@pedro-salgueiro
thanks for your input, maybe that is the culprit

I did a simple test, with a IMHO simple prompt

create 150 c# files, each file shoud contain a class with a random name, each class should have a print method which prints the class name

and it stops after creating about 20 to 25 files, it isnt always the same amount of files

since it may be related to visual studio, I created this issue:

https://developercommunity.visualstudio.com/t/Why-does-copilot-stop-in-the-middle-of-t/10958679

in the next couple of hours I will try if the same issue does happen with different AI Provider, as I said earlier currently using Claude Sonnet 4

edit:
update tried it with gpt 5, had to change prompt to

create 150 c# files, each file shoud contain a class with a random name, each class should have a print method which prints the class name
dont use a script, create each file indepentadly and use firiendlier names not guids

as to not get a powershell script which tried to generate all files with guids in 1 step
gpt 5 than completed the task

also tried to use gemini 2.5 pro, but I am facing performance issue and it keeps stuck again and again, so maybe I give it some other day a try

[jgwinner]

PMFJI - I've been running into the same bug with Visual Studio 2026/Preview. It's up to date.

Ironically, I can't duplicate it right now, but here's a screenshot example:

There were no errors showing in the output. It's just hung ... clicking on the 'stop' button or pressing 'enter' doesn't do anything. The hand up/down vote icons don't show up either.

I don't think it's a rate limit, as I restarted VS and it started working (of course). I've had this happen a lot now; restarting VS every time is getting tiring.

If it IS a rate limit, I agree that there should be some kind of warning. I'll admit I'm using a free version.

[verdantburrito]

Running into the same problem here, on the latest release version of Visual Studio 2026 with a Copilot Pro (paid) plan subscription. The behavior doesn't appear to be model-specific; I mostly used Claude Sonnet 4.5, but also tried Claude Opus 4.5 for 30-60 minutes, & I constantly had to tell it to continue in both cases.

If it's a rate-limiting problem, then it's definitely a bug because there's no error message reporting it as such. And FWIW I got a legitimate rate-limited error message last night when I exceeded the burst message threshold. After waiting an hour before retrying, that problem went away; this other problem with having to constantly tell the agent "Continue" has been occurring for at least a day now.

[JohanAR]

It seems like sometimes when subagents ask for permission, nothing is written in the chat. Clicking the arrow to expand the subagent's work shows its initial message and nothing more. Only when I back out to the agent sessions list (which btw. defaulted to hiding "read" chats so it looked empty) and reopen the chat I get the full subagent history and the question it's waiting for.

[Abhirai2006]

Hello 👋

This occurs typically due to the fact that Copilot splits your request into several steps or fails to retain the context during work, particularly in large-scale projects.

Reasons 📢

• Large-scale project ➡️ reaching token context limits
• Interruption between requests (a feedback prompt like "Was this helpful?")
• Session/context timeout

Possible Fixes 💻

• Try splitting up your requests into separate parts rather than posing a single massive prompt
• Manually define your files using #filename syntax
• Make your prompt concise and specific
• Try restarting the chat/conversation

Caveat 🛑

The current behavior of Copilot isn't entirely a bug but a known feature; sometimes, Copilot will pause and wait for user input.

[Gecko51]

@pedro-salgueiro identified the actual cause earlier in the thread. In VS Code, there's a setting called "Agent: Max Requests" (github.copilot.chat.agent.maxRequests) that controls how many tool calls the agent can make per conversation turn before stopping to ask for confirmation. When it hits the limit, you get that pause.

The problem: Visual Studio doesn't expose this setting. No registry key, no options panel entry. The Copilot agent in Visual Studio is still behind VS Code in configurability.

Your 150-file test confirms it's not rate limiting. A rate limit would give you an error message or a retry delay. The fact that typing "continue" immediately resumes the work with no wait proves the model is perfectly fine, it's just hitting the per-turn request cap.

For now, the realistic workarounds:

• Ask for smaller batches ("create 20 files with this structure, I'll continue") instead of one large task. Not ideal, but it gets work done without the interruptions.
• Keep your Visual Studio Developer Community thread alive. This is a known gap between VS and VS Code Copilot behavior and needs user votes to get prioritized.

On GPT-5 completing the 150-file task where Claude Sonnet didn't: probably comes down to how the two models handle agent planning and batching operations, not a fundamental capability difference. Worth trying a prompt that explicitly tells the agent to group file creation into batches rather than handling files one at a time.