GitHub Community
GitHub Copilot Coding Agent Has Arrived! 🚀 #159068
Replies: 40 comments 61 replies
-
|
Where can we see this in action? Any real Issues to PR examples? |
Beta Was this translation helpful? Give feedback.
-
|
Hi @cquanu Here's our highlight videos: |
Beta Was this translation helpful? Give feedback.
-
|
Rather than a video, I think that it would be helpful to share a real time pull request and more so when the repository in the video is private |
Beta Was this translation helpful? Give feedback.
-
|
It would have been nice to see the real time action when they decided to launch the product. |
Beta Was this translation helpful? Give feedback.
-
|
Here you see it in action: https://www.reddit.com/r/ExperiencedDevs/comments/1krttqo/my_new_hobby_watching_ai_slowly_drive_microsoft/ |
Beta Was this translation helpful? Give feedback.
-
|
FYI if you upgraded or are already on Copilot Pro+ and were confused because Co-pilot was not popping up to be assigned tasks, you need to go to your Co-pilot settings where a new "Copilot agent" tab is available then give it access to all or some of your repositories. |
Beta Was this translation helpful? Give feedback.
-
|
If I book Copilot Pro+, can I use it also in organization repositories? |
Beta Was this translation helpful? Give feedback.
-
|
As far as I can tell you can. I haven't actually tried it yet tho. But it shows up. |
Beta Was this translation helpful? Give feedback.
-
|
It shows up also with regular Copilot :) |
Beta Was this translation helpful? Give feedback.
-
|
Yes, you should be able to so long as the organization repositories have been enabled to use the agent. |
Beta Was this translation helpful? Give feedback.
-
|
Really cool! Might implement in my codebases :) |
Beta Was this translation helpful? Give feedback.
-
|
What model is it using behind the scenes? |
Beta Was this translation helpful? Give feedback.
-
|
Hi @fishcharlie ! Coding agent currently leverages Claude Sonnet 3.7. |
Beta Was this translation helpful? Give feedback.
-
|
That explains why it consumes premium requests. |
Beta Was this translation helpful? Give feedback.
-
|
Pretty cool. Used it a few times so far. Needs a bit of handholding, but that's expected. It's pretty slow, and I think that's because it has to "learn" the repository each time it spins up for a new issue. Maybe adding the guidelines file will speed that up. The biggest issue I have right now is that it downgrades java projects to match the version it has installed. Will need to do this I suppose https://docs.github.com/en/enterprise-cloud@latest/copilot/customizing-copilot/customizing-the-development-environment-for-copilot-coding-agent |
Beta Was this translation helpful? Give feedback.
-
|
Yes, that is recommended. Occasionally the model lacks the latest knowledge, so the agent will benefit from any custom set up steps that you add to download the correct version. Otherwise, you can also leave a comment on your PR telling Copilot that there is a new version and that should be leveraged instead. |
Beta Was this translation helpful? Give feedback.
-
|
This is too much. The current direction of AI use is not good. We don't need Github or Copilot to do software engineer's job. Just because people put code to your repos, it doesn't mean you own them. They have their own licenses. You shouldn't steal and use people's code to create new software. They are put here so that people can contribute to those projects. |
Beta Was this translation helpful? Give feedback.
-
|
Session Time Limits, Progress Saving, and Duration Increase Regarding the agent's session duration, there are a couple of points that would significantly improve the workflow. Firstly, it seems the current sessions are subject to a time limit of 60 minutes. If the agent doesn't complete its task and commit the code within this timeframe, the entire session's work appears to be lost. It would be extremely beneficial if the agent could be made aware of this time limit and, if it cannot complete the task, at least commit its partial progress to a branch before the session expires. This would allow users to continue the work manually or reassign the agent from that saved state, preventing the loss of valuable processing time and effort. Secondly, for more complex issues or larger repositories requiring extensive analysis and changes, the current 60-minute limit can be quite restrictive. Could the maximum session time limit for the Copilot Coding Agent be increased? Perhaps extending it up to 180 minutes would allow the agent to successfully complete more substantial tasks in a single run, reducing the need to break down issues into smaller parts or repeatedly reassign. Combining these two ideas – increasing the potential duration and adding a mechanism to save partial progress before timeout – would greatly enhance the reliability and usefulness of the agent for a wider range of tasks. |
Beta Was this translation helpful? Give feedback.
-
|
+1 @Akash1134 Any update on this? |
Beta Was this translation helpful? Give feedback.
-
|
Hi @abhi-parate-st3 thanks for the feedback! The time limit of 60 minutes is intentional to help prevent abuse. Regarding your first point, this is something that we are actively looking into. Agreed that this would be a much better experience. On your second point - the maximum session time limit cannot be increased at the moment. Maxing out the session limit is usually an indicator that something is awry. Typical runs are tens of minutes rather than hours. With that being said, we'll keep a pulse on this and reevaluate with the improvements coming from your first point. |
Beta Was this translation helpful? Give feedback.
-
|
Self-Hosted Runner Support For organizations with specific security requirements or those dealing with larger/more complex codebases, support for self-hosted runners would be a valuable addition. This would allow us to run the Copilot Coding Agent tasks on our own infrastructure, potentially enabling longer task durations, faster execution on dedicated hardware, and meeting compliance needs. |
Beta Was this translation helpful? Give feedback.
-
|
+1 to this I have been working in HIPAA and PCI-DSS regulated environment and this could be a requirement in future. |
Beta Was this translation helpful? Give feedback.
-
|
+1 @Akash1134 This is a must to scale this for small companies. Any ETA or work is in progress or not? Please confirm so we can think of other similar services like codex and google jules |
Beta Was this translation helpful? Give feedback.
-
|
We do not currently support self-hosted runners. I've opened a feature request ticket to track this. Moving forward if anyone else has organization/company specific requirements related to this request, please drop them below ⬇️ |
Beta Was this translation helpful? Give feedback.
-
|
in my organization, due to the highly confidential nature of our work, we are strictly limited to using self-hosted runners. While I am very excited about the potential of this product, its current lack of support for self-hosted runners unfortunately makes it completely unusable for us. This is a significant limitation and a critical requirement for organizations with stringent security and compliance needs. I urge you to seriously consider prioritizing the implementation of self-hosted runner support. |
Beta Was this translation helpful? Give feedback.
-
|
We have self hosted registries for multiple languages (pypy,npm,and nuget) and containers (docker, oci) that all exist within our corporate network. We have been discussing how we might allow access for copilot-agent but the answer has always gone back to. We already support this on self-hosted runners why reinvent the wheel. We have seen in the past with GHAS scanner it supports self-hosted runners now. It would be great to have the option for copilot-agent as well. |
Beta Was this translation helpful? Give feedback.
-
|
Larger GitHub-Hosted Runner Support for Free Organizations Currently, it seems access to larger runners might be limited based on organization plans. Could there be an option to use larger GitHub-hosted runners for Copilot Coding Agent tasks even within free organizations, perhaps by consuming more GitHub Actions minutes or premium requests? This would help tackle more demanding tasks that might time out or fail on standard runners, making the agent more useful for a wider range of projects. |
Beta Was this translation helpful? Give feedback.
-
|
+1 @Akash1134 This feature can help as of now to scale with a coding agent for any org. Any ETA? |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for sharing this feedback. I'll share this with the GitHub Actions team for consideration. |
Beta Was this translation helpful? Give feedback.
-
|
Availability for Copilot Pro Users The announcement mentions availability for Copilot Pro+ and Enterprise subscribers. Given that the agent will consume Copilot premium requests which Pro users also have access to, could this feature be made available to standard Copilot Pro subscribers as well? Many individual developers and smaller teams using Copilot Pro could greatly benefit from the agent's capabilities to manage personal projects and contributions. |
Beta Was this translation helpful? Give feedback.
-
|
Yes, our plan is to eventually roll out to additional SKUs such as Copilot Pro. There is no timeline to share just yet. Once this is available, we will make an announcement here as well as in a changelog. Stay tuned! |
Beta Was this translation helpful? Give feedback.
-
|
That will be awesome! |
Beta Was this translation helpful? Give feedback.
-
|
Hi, I saw the announcement that the Does the use of the word 'now' suggest that there are plans for the Copilot Coding Agent to become available for other plans, such as Copilot Business or other individual plans, in the future? |
Beta Was this translation helpful? Give feedback.
-
|
Yes exactly! We will continue to roll out to other plans in the coming months. |
Beta Was this translation helpful? Give feedback.
-
|
Does this replace the copilot-workspace experiment? |
Beta Was this translation helpful? Give feedback.
-
|
Same question. Is https://copilot-workspace.githubnext.com/ still a thing or it will be abandon |
Beta Was this translation helpful? Give feedback.
-
|
They are similar - Copilot Workspace takes a task, implement it and create a PR - but it has a linear, highly structured flow, and isn't deeply integrated into the GitHub tools that developers already use like issues and PRs. With Copilot coding agent, we're taking all of the great work on Copilot Workspace, and all the learnings and feedback from that project, and integrating it more deeply into GitHub and really leveraging the capabilities of 2025's models, which allow the agent to be more fluid, asynchronous and autonomous. |
Beta Was this translation helpful? Give feedback.
-
|
I'm curious about how the coding agent will handle more complex project structures. For example, if we have three different parent feature issues, each with ten story issues as sub-issues, how will assigning to those parent issues work for all their respective child issues? And can Copilot figure out the issue implementation sequence across these related issues by itself, or will we need to define that order? |
Beta Was this translation helpful? Give feedback.
-
|
Currently, when assigning the agent to a parent issue, the agent will only act on that specific issue, and does not take in context of its sub issues. This is something that we intend to more deeply explore in the next few months. |
Beta Was this translation helpful? Give feedback.
-
|
alright thanks for clarity and reply. |
Beta Was this translation helpful? Give feedback.
-
|
I'm curious to hear more about how this feature works if you make changes to the underlying codebase. For example if I go in and make a few commits to the PR that Copilot makes, then leave comments as a review, will that cause problems? Obviously there is a point at which this probably becomes unreliable. But curious what factors I should be thinking about here and how Copilot handles that. |
Beta Was this translation helpful? Give feedback.
-
|
@nhu-do |
Beta Was this translation helpful? Give feedback.
-
|
@nozaq There's a bug right now that prevents re-assignment. We're tracking that as a P0 bug to get this workaround back up. Appreciate your patience and testing out the agent! |
Beta Was this translation helpful? Give feedback.
-
|
@nhu-do Understood. Thanks for the prompt reply! I hope it gets resolved soon 🙂 |
Beta Was this translation helpful? Give feedback.
-
|
@nhu-do It seems that this reassignment issue still hasn't been resolved. |
Beta Was this translation helpful? Give feedback.
-
|
You're correct - the fix has not been shipped yet due to competing priorities. Thanks for opening that issue 🙏🏼 |
Beta Was this translation helpful? Give feedback.
-
|
There seems to be a 120 s maximum run-time setting somewhere but I cannot find it ;( Of course yarn takes ages to run, so .... |
Beta Was this translation helpful? Give feedback.
-
|
Workaround is to run yarn (yarn install) as part of the setup. |
Beta Was this translation helpful? Give feedback.
-
|
Now that premium requests are being tracked, these timeouts are even less palatable as it burns though our allowances for no benefit. |
Beta Was this translation helpful? Give feedback.
-
|
Congrats on the release of GitHub Copilot Coding Agent! 👍 After assigning Copilot to an issue, I would like notifications to be sent to the user who assigned Copilot at the following timings:
|
Beta Was this translation helpful? Give feedback.
-
|
Hello. I just subscribed to copilot pro+. I still can't assign issues to copilot:
|
Beta Was this translation helpful? Give feedback.
-
|
It would be great if there's a free tier for public repos with data sharing, similar to OpenAI’s data sharing program or Anthropic’s developer partner program. |
Beta Was this translation helpful? Give feedback.
-
|
I've seen some interesting behaviour when you give it an image to explain a concept / show a visual defect.
|
Beta Was this translation helpful? Give feedback.
-
Am impressed with Copilot's graduate knowledge (~100 messages, just the first is about personal repo, after which the discussion moves to calculus + quantum compute + neuroscience). |
Beta Was this translation helpful? Give feedback.
-
|
Seems the Copilot coding agent policy is missing in our Enterprise environment, any idea why? |
Beta Was this translation helpful? Give feedback.
-
|
Support confirmed with me that you as a Enterprise Admin need a Copilot Enterprise License attached to you before you will see these permissions. I find this wild but it did resolve the issue. |
Beta Was this translation helpful? Give feedback.
-
|
Same for me |
Beta Was this translation helpful? Give feedback.
-
|
Are there any information when (or if at all?) this will be available for managed enterprise accounts? |
Beta Was this translation helpful? Give feedback.
-
|
Yes, coding agent is available for managed enterprise accounts. The only limitation there is it is not available for namespace repositories owned by such accounts as the agent requires GitHub hosted runners. See more here. |
Beta Was this translation helpful? Give feedback.
-
|
I've signed up for Pro+ a few days ago specifically for this feature. I used it for 2-3 days, on and off, it's not like I kept the agent working 24/7, and it just burned through my quota of premium requests and now the feature is just disabled. Very disapointed. There is no control over which model Coding Agent uses, so if it decides to use GPT-4.5 it will burn through those alloted 1500 monthly premium requests in no time as each GPT-4.5 request counts for 50x premium requests. There really needs to be a way to specify which model you want the coding agent to use. Furthermore, that the feature would just turn off completely, after 2-3 days of mild usage is just insane. It should atleast revert to using some of the free models or something. There is no way I'm going to pay the 0.04 per premium request if you don't give me the ability to select which model it will use. So far, a typical Coding Agent session runs for about 20-25 minutes and uses up anywhere between 25-75 premium requests + the associate github actions cost. If someone were to use this to its full potential and keep the Coding Agent relatively busy all day, every day, this would end up costing a small fortune. For some projects, it might make sense... but I'd still want control over what models (and therefore how costly) each Coding Agent session is using. |
Beta Was this translation helpful? Give feedback.
-
|
100% agree. The cost aspect is a major concern. There are so many things they could do to improve that. But right now it's unsustainable. One note however, Coding Agent has a static premium request multiplier of 1: So it doesn't really matter to consumers what model it uses under the hood (in terms of price). But if you could select 4o or 4.1, GitHub would likely change that policy and reduce the cost. I do question if 4o or 4.1 could really perform well in this environment. But at least it'd be free. Regardless, they need to improve the pricing structure. |
Beta Was this translation helpful? Give feedback.
-
Oh, that's good to know. I hadn't seen that footnote. But in a way that makes it worse. It means I have no hope of reducing the premium request my Coding Agent sessions are taking as they already just count as one. Darn! I have to say this is a bit disheartening. The tech is amazing and a litteral game changer. Those that can afford it and leverage it will be more than 10x more productive than the ones that can't. Where does that leave open source software? Side projects? Freelancers? Bootstrapped startups? At this price point, it feels like in a very near future, the only orgs producing code will be well-funded mega-corps that hire a handful or hyper productive "project manager/architects" that are good at orchestrating the work of asynchronous coding agents. Everyone else won't be able to afford the tools necessary to produce software fast enough to compete. Things are moving so fast that people don't realize yet the difference between something like GH Copilot Coding Agent and the in-IDE Copilot Chat agent. The fact that it runs in a sandbox and doesn't have to ask you permission for every command it wants to run is the real difference. Come to think of it, I don't see why we couldn't have that experience locally. |
Beta Was this translation helpful? Give feedback.
-
|
@snowfrogdev Couldn't agree more. I referenced your comment in this thread: copilot-coding-agent/user-feedback#48 (comment). I hope @nhu-do and the rest of the GitHub team will take our comments to heart. The technology is great. But we just need better pricing. I'm not saying GitHub should give this away for free. But I also don't think their pricing is very competitive right now. |
Beta Was this translation helpful? Give feedback.
-
|
Agree better pricing only can empower adoption of copilot coding agent. |
Beta Was this translation helpful? Give feedback.
-
|
👋🏼 Thanks for your feedback everyone. We heard you loud and clear, and were working to make the pricing model more affordable and predictable to enable your use of coding agent. Today, we're excited to announce an updated pricing model, where 1 session = 1 premium request. Check out our changelog! 🚀 |
Beta Was this translation helpful? Give feedback.
-
|
It seems the Coding Agent is now available on the Copilot Business plan, but it's very difficult for us to adopt. The Team plan only includes 3,000 free minutes for GitHub Actions for the entire organization, whereas the Pro plan gives a single user 3,000 minutes. To put it simply, even if one task takes 10 minutes, that's only 300 runs in total (3,000 ÷ 10). If we have 50 members in our organization, each person could only use it 6 times to stay within the free limit. We might consider paying for it if it proves useful in the future, but with this limit, it's difficult to even test it, especially considering our existing usage of GitHub Actions. Could you either expand the free tier for GitHub Actions or allow the use of the Coding Agent without consuming GitHub Actions minutes? |
Beta Was this translation helpful? Give feedback.
-
|
Is there something special that needs to be done in order for the Copilot Agent to have access to read the GHA workflow logs? I would expect it to watch the PR for failing checks and access the logs to see if further action is required without my having to manually prompt it. Even when prompting it to fix the failing check, it did not appear to fetch the logs even when I provided a similar prompt that is issued when asking Copilot to explain the error from the failing check's menu option. |
Beta Was this translation helpful? Give feedback.
-
|
Hey there, |
Beta Was this translation helpful? Give feedback.
-
|
Has anyone had any success with the default browser that the coding agent is supposed to have access to? When it tries to view an svg file in the browser it consistently gets the following error: The logs show it using this input: The agent also tried creating a tmp html page with links to the svg image. If fails to view this page and gets the same error. |
Beta Was this translation helpful? Give feedback.
-
|
Never had that issue. My coding agent was able to launch dev server and even take a screenshot for the PR description. See here: |
Beta Was this translation helpful? Give feedback.
-
Are there any plans to respect this setting for future copilot features? |
Beta Was this translation helpful? Give feedback.
-
|
does github copilot coding agent is going to respect instructions.md all the time like i am asking because in agent mode while generating new file and code it is not working in vs code. Thanks |
Beta Was this translation helpful? Give feedback.
and others
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
🚀 Say Goodbye to Backlog Burnout
Let's be honest – we've all felt that sinking feeling when looking at a growing pile of issues. Technical debt creeping up? Features waiting to be implemented? Bugs that need squashing?
Enter
Copilot Coding Agent – your new AI teammate that doesn't need coffee breaks or complain about fixing that legacy code everyone avoids! Let's roll!
Read the detailed post from the GitHub CEO: GitHub Copilot: Meet the new coding agent ✨
💪 How It Works (It's Ridiculously Simple)
Just assign issues to Copilot the same way you'd assign them to any team member – through:
...or by simply asking Copilot in chat!
Then watch the magic 🪄 happen! Copilot takes it from there:
🌟 Perfect for Those "Important But Not Urgent" Tasks
Copilot Coding Agent shines with:
The best part? You can assign multiple issues simultaneously and let Copilot multitask while you focus on the complex, creative work that requires your human touch.
🔑 Getting Started
Copilot Coding Agent is available now for:
Using the agent will consume GitHub Actions minutes and Copilot premium requests from your plan's included entitlements.
Ready to supercharge your productivity? Check out our comprehensive GitHub Docs for pro tips on getting the best results!
💭 We Want Your Feedback!
This is just the beginning of an exciting journey. Have you tried the Coding Agent yet? What tasks are you delegating? Any feature requests or questions?
In addition to this discussion post, if you encounter any bugs or have feature requests, please file them in our user feedback repository.
Share your thoughts, wins, and even your wildest ideas below! Let's shape the future of AI-human collaboration together!
🚀💻✨
Beta Was this translation helpful? Give feedback.
All reactions