What happened: Github cloud agents write the PRs independently
Github has announced a new major capacity for Copilot:
Github unveiled a significant upgrade to Copilot during Microsoft Build 2025: a cloud -based coding agent capable of writing and iterating traction requests directly in Github. While previous co -pilot experiences focused on editor assistance via VSCODE, this new agent works asynchronously in the cloud, taking advantage of an infrastructure similar to Github actions.
Developers can now affect tasks directly via GitHub or VSCODE:
> @github Open a pull request to refactor this query generator into its own class
The co -pilot agent manages the tasks independently: creation of branches, iteration on PRS depending on the comments of code revision and updating of commitments until the work is accepted – all without touching protected branches. Above all, existing CI pipelines, branch protections and examine the workflows remain intact, guaranteeing “confidence by design”.
With the addition of MCP, developers can even grant the agent access to external tools and data by configuring servers directly in the framework of the repository. This sums up us with the actual implementation and allows us to focus on the description of the tasks. This is aligned with codex agents and others providing agents to execute and coding a calculation.
He also resonated with the wider movement towards headless agents – those who execute independently, perform tasks and report when the work is ready to be revised.
Community reactions: fear, prudence, fears and questions.
Personally, the use of agent Github Copilot was a pleasant experience when he asked to perform low level tasks. I had a “wow” moment when I carried out typical GitHub actions directly inside the PR (or via the VSCODE cat bar). It was impressive and fast and the experience of the developer felt there.
By digging in the reaction of the community, we found the coding agent of Copilot showing a mixture of excitation, curiosity and healthy skepticism. “Time saving” was common praise. Users liked the public relations project and the agent's newspaper allowed them to see exactly what was going on and that they have kept the choice of when to merge. Similar to my experience, the users have shared, they found Copilot to work well with low level tasks.
To which the coding agent of Co -Pilot of Github replied:
Part of the community also shared a fear that by discharging coding towards AI, human work could be transformed into concierge surveillance: the Detailed Jira tickets all day and the code generated by Ai-Généré. In addition, users have expressed themselves by what it means for junior developers. The vision of the CEO of Github on this subject is:
At a time when some claim that more AI means fewer opportunities for entry -level developers, I believe that the opposite is true. There was never more exciting time to join our industry.
Based on this concern, security and political issues have also been raised. For example, could the agent inadvertently expose sensitive information inadvertently, like including a secret key in an PR? Adding to complexity, some developers have stressed that the advantages of the use of AI agents do not come for free. You only harvest the awards if you are investing efforts in advance; For example, write a good description of a problem with clear acceptance criteria.
Do you take advantage of being informed of the indigenous development space of AI (ANind)? I help build the AIND community, where we cover the most relevant news developers. For busy people, we send a clean newsletter to a pager by more than 7,000 breakers AI each week – Free and directly in your reception box!
The Aind Take: Summary of grunts and the development of new habits
Github evolves towards a development platform improved by AI. Like the historical growth of marketplace actions, we can expect a peak in the tools in co -pilot extensions, making the co -pilot more powerful and covering more aspects of the development workflow.
By pulling an analogy of history, it seems close to the Devops revolution or the transition to Cloud / Server IT: the banal infrastructure work is abstract, allowing developers to focus on higher level logic. Likewise, the co -pilot agent can abstraction a growing piece.
It is also a new competence: to invite and supervise AI in coding. Just like CI / CD and Cloud have led to new roles, AI agents could lead to new roles like “AI orchestration engineer”. I suspect that we will see a-native startups experimenting more and more with the fact that several agents collaborate.
One use case in which we believe is the use of agents dependent on tasks: an agent could generate code while another examines it, or an agent specializes in front and another in the back-end, by coordinating through problems and PR. Github's infrastructure could support this type of multi-agent workflow, especially since MCP allows you to chain different AI services.
From the point of view of a developer, to prepare for this future centered on AI, there are a few concrete steps that we recommend when reading with the Copilot agent. First, writing tasks / clear and detailed public relations / public relations with acceptance criteria will become a precious competence. It is a good practice even for human collaborators, but for AI, it is very relevant. This helps to develop communication muscle with machines on the intention of the code.
Second, invest in tests. The prospect of an AI agent who relies on tests to know that he has not broken things is a strong incentive. High test coverage can transform the agent into a reliable contributor, while the low coverage makes it a serious risk. The strengthening of your automated tests and your CI pipelines can help position your projects to benefit in more detail from the involvement of AI.
The end game is to become comfortable letting the AI ​​write something and guide it / refine it. In this new era, the developers who flourish will be those who will adopt the development focused on specifications, guiding the AI ​​with precision, shaping code and validating it with tests.
