Why this update matters
AI coding assistants have moved beyond answering questions. The real question now is whether they can complete a task, follow the right path, and avoid creating extra work for the developer.
Microsoft’s latest Copilot Chat upgrade in Visual Studio targets exactly that gap. Starting with Visual Studio 2026 version 18.12, Copilot runs on the Microsoft Copilot SDK developed by GitHub, the same technology behind GitHub Copilot CLI, the GitHub Copilot app, and VS Code.
The result is not a flashy new model. It is a more practical improvement: Copilot completes more coding tasks, gives shorter and more focused answers, and makes it easier to see what it is doing while it works.
For development teams, that can mean less time rewriting prompts, checking output, and fixing avoidable mistakes.
What Microsoft changed
The upgrade changes the engine behind Copilot Chat, not necessarily the underlying AI models. Microsoft says the new experience uses the same models as before, but organizes and executes the work more effectively.
That distinction matters. A model can be highly capable, but still produce a response that is too broad, loses context, or stops before the task is genuinely complete. The new Copilot harness is designed to keep working until the task is finished, rather than treating a partial answer as a final result.
Microsoft describes the update as a move toward “less chatter, more quality results,” especially for larger tasks.
The productivity test
Microsoft tested the old and new Copilot experiences across 157 coding tasks drawn from real C# projects. The projects covered .NET Framework and .NET Core, Azure and Aspire, and WinForms and desktop applications.
A task was only counted as resolved when failing tests passed without breaking tests that already passed. That is a useful standard because it measures working software, not just a plausible-looking answer.
| Measure | Previous Copilot experience | Updated Copilot experience |
|---|---|---|
| Tasks tested | 157 | 157 |
| Tasks resolved | 103 | 124 |
| Additional tasks resolved | – | Up to 21 |
| Approximate improvement | – | About 20% |
The updated experience resolved 124 tasks, compared with 103 for the previous version. That is up to 21 additional completed tasks, or roughly a 20% improvement.
Microsoft also says the new engine performed better across the models it tested, and that it tested C++, Python, and other languages beyond C#.
Why first-attempt accuracy matters
Developers do not usually judge an AI assistant by how confidently it explains a concept. They judge it by whether it saves time.
If Copilot suggests a fix, the developer still needs to review it, run tests, check edge cases, and make sure it fits the wider codebase. If the first attempt is wrong, that review becomes a correction cycle: clarify the prompt, regenerate the code, inspect the changes, and test again.
The difference between a useful response and an almost-useful response can easily be several minutes. Multiply that across a day of coding work and the cost becomes significant.
This is why the 20% improvement is more meaningful than it may first appear. It suggests fewer restarts, fewer unnecessary explanations, and more tasks moving from “in progress” to “done.”
A more controlled workflow
The update also changes how developers interact with Copilot during a task.
Previously, a developer might need to wait for Copilot to finish before changing direction. Now, they can type a new instruction while Copilot is still working. The message appears as pending, and the developer can remove it before Copilot picks it up.
This is a small feature with a large practical impact. Software work is rarely linear. A developer may start by asking Copilot to update a component, then realise the test approach needs to change, or spot a dependency issue halfway through.
Instead of cancelling the task and starting again, the developer can steer it in real time.
More visibility, less guesswork
Another improvement is transparency. Copilot now shows the results of each action, not only the steps it attempted. Developers can expand a step, view the result, open it in the editor, or copy the exact command Copilot ran.
That makes the assistant easier to audit. Rather than accepting a black-box answer, developers can follow the reasoning and verify the work.
When Copilot delegates part of a task to a subagent, it displays a dedicated card showing the model in use and every step the subagent takes.
For teams, this matters because accountability is part of productivity. A developer needs to know what changed, why it changed, and whether the change can be trusted before merging it into a project.
Working beyond one solution
The updated Copilot can also work with folders outside the current Visual Studio solution. Developers can point it to a folder on their machine from the chat header or attach whole folders as context.
That is useful for real-world projects, where relevant code, documentation, configuration files, and test assets are often spread across several locations.
It also supports the use of built-in agents. With Agent selected, developers can type / to call one of the available agents for a more specialised task.
Consistency across Microsoft tools
Because Visual Studio now shares the same Copilot technology as VS Code, GitHub Copilot CLI, and the GitHub Copilot app, the experience should feel more consistent across tools.
For a developer moving between an IDE, terminal, and code editor, consistency reduces friction. The same request should not require a completely different approach depending on where it is made.
For businesses, this can also simplify adoption. Teams do not need to learn a separate AI workflow for every development surface.
What this means for teams
This update is best understood as a productivity improvement, not a replacement for engineering judgement.
Copilot may be better at completing tasks, but developers still need to:
- Review generated code before merging it
- Maintain clear tests and validation processes
- Provide relevant context, constraints, and coding standards
- Treat AI output as a draft, not a final decision
- Monitor security, performance, and maintainability
The biggest gains will likely come when teams use Copilot for well-defined work: fixing a failing test, updating a component, refactoring a known pattern, writing documentation, or investigating a bug.
The practical takeaway
Microsoft’s Copilot upgrade is important because it focuses on the part of AI coding that actually affects delivery: completing work correctly with less rework.
The updated Copilot Chat in Visual Studio resolves more tasks, provides more focused responses, allows mid-task steering, and gives developers clearer visibility into what it is doing.
For developers, the next step is simple: update to Visual Studio 2026 version 18.12, open Copilot Chat, and test it against a real task from your backlog.
The value of AI in software development will not come from replacing developers. It will come from reducing the time developers spend correcting, clarifying, and cleaning up work that should have been done properly the first time.
