At Build 2025, Microsoft introduced a new upgrade to GitHub Copilot called the Copilot Agent.
Unlike previous versions that only suggested code, this new version is built to take on full coding tasks from start to handoff - like fixing bugs, refactoring code, and opening pull requests.
It’s powered by Anthropic’s Claude 3.7 Sonnet model and is currently in preview, with Microsoft gathering feedback before a broader rollout.
What the Company Is Saying
Microsoft’s positioning is straightforward: this tool can now be given a task - such as "fix this error" - and it will:
-
Make the code change
-
Write the commit message
-
Open a draft pull request
-
Summarise what it did
The developer still reviews and approves the changes, but the execution and draft handoff is now handled by AI.
What That Means (In Human Words)
This isn’t a smarter autocomplete.
This is Microsoft building an assistant that:
-
Understands where the bug is
-
Applies a fix across the codebase
-
Prepares everything for team review
-
And waits for you to merge
Instead of working line by line, it works in tasks.
You assign. It acts. You approve.
What It Can Actually Do
Here’s what the Copilot Agent is designed to handle (according to Microsoft):
-
Navigate large codebases
-
Fix known bugs
-
Add or modify features
-
Refactor older logic
-
Write documentation
-
Draft pull requests with summaries
-
Handle repetitive cleanup and standards
-
Work directly in VS Code and GitHub environments
What’s Still Missing?
As of now, there’s no public detail on:
-
How it performs in real-world, messy codebases
-
Whether developers will trust its work in critical systems
-
How teams can audit or trace what it changed
-
Whether it requires code to be well-documented or follow specific standards
-
How access and permissions are controlled in shared repos
Right now, it’s in preview. This is still a test phase.
🧊 Bottom Line: Cost, Availability, and What You Need to Know
✔️ Feature: GitHub Copilot Agent
✔️ What it does: Completes full code tasks (not just suggestions)
⚙️ Powered by: Claude 3.7 Sonnet (Anthropic)
📍 Where it works: GitHub + VS Code
🧪 Status: Preview (limited rollout)
✅ Requires: Human approval before merging changes
📢 Announced at: Microsoft Build 2025
Want to Read More?
🔗 Watch the announcement (YouTube)
📰 No official blog post yet
❄️ Frozen Light Team Perspective
The main reason bugs exist isn’t mystery - it’s human drift.
We forget things.
We skip steps.
We aren’t great at repeating the same fix the same way every time.
Most bugs don’t need genius. They need consistency.
And that’s exactly where AI fits in.
You can wrap this up in branding - say Copilot acts like a teammate or takes initiative.
But the truth is simpler:
👉 This is about automation at scale.
👉 And AI is at its best when repetition is the problem.
Bug fixing often is repetition:
-
The fix has been done before
-
The error looks like others we’ve seen
-
The pattern is clear
-
The work is tedious
Will this new Copilot Agent do a good job?
We don’t know.
Will developers trust it fully or only in certain cases?
Still we don't know.
Will it need clean, well-documented code to work?
Maybe.
But here’s what we do know:
If this works, it will streamline the one part of development almost everyone dreads - fixing bugs.
And that’s a good enough reason to try.
