GitHub Universe 2025 ended a few weeks ago, but there was a ton of cool announcements and some stuff is still yet to come (I'm waiting for the end of the year 👀). I want to share quickly what I learned from watching the sessions and my opinion on some topics.
Here is a quick recap too:
Table of Contents
TL;DR
- Agent HQ is a big focus: GitHub becomes mission control for all your coding agents (Anthropic, OpenAI, Google, xAI, and more) with unified task management, granular security controls — all included in your Copilot subscription.
- Copilot leveled up significantly: Code Quality, custom agents and code review improvements. There were significant upgrades to Copilot and it's definitely a lot better!
- Copilot metrics dashboard: GitHub launched usage dashboards and APIs showing acceptance rates, engagement, Lines of Code (LoC) related metrics and more
Agent HQ - Mission Control
Here is the promotional video on Agent HQ:
Sure, it looks cool. Many demos during day 1, from Anthropic and OpenAI looked good, but demos always look good 😅. They are pushing to give us tools to orchestrate a fleet of specialized agents, then monitor and iterate on their work while the agents work in parallel.
A great new addition is the ability to steer the agent while it's working!! Super awesome, since we could even send a large PRD in that chat session, to introduce new requirements or context to the agent. Previously, we only "communicated" with Copilot coding agent through comments in the issue/PR.
They also want the Agent HQ to be the place where you define all security controls regarding agents, for example, an agent runs with a locked-down GitHub token that limits exactly what it can do. We get audit logs, the firewall copilot has right now, and other cool stuff.
Here is another big promise from GitHub: Over the coming months, coding agents from Anthropic, OpenAI, Google, Cognition, xAI, and more will become available directly within GitHub as part of your paid GitHub Copilot subscription.
If this is true for all paid plans (not just Pro+ and Enterprise), it would be fantastic. We can pay $10 for Copilot Pro and get most agents, plus most frontier models, with some being free (e.g. GPT-5 mini). Of course, there would be downsides, for example, you may have Claude code, but we probably lose access to claude.ai and Claude code CLI. Will this GitHub integration be 100% the same as the CLI? Will there be some parts of the integration open-source? Will AGENTS.md work on them all or not (e.g. currently Claude code doesn't support it)? Will we have the deep research tool in Claude code agent like we do in claude.ai? Will rate limits be the same? Right now I'm not sure marketing and sales teams are only selling dreams. GitHub wants to be the hub of everything agent related, they are partnering with everyone, but I'm skeptical by nature 😅 . I'll believe it when I see it.
Still, very cool announcement!
GitHub Code Quality
New feature in preview to ensure code quality on the repository. This feature focus on maintainability and reliability of the code right now, then in future releases add test coverage and "AI era challenges". Plus we can block PRs that don't meet code quality standards.
There is another tab now for AI findings where we can see suggestions to recently modified code. I mean if we ran code review agents at PR time, this feature wouldn't be that helpful, but if it does find problems we can assign the Copilot coding agent to fix them all.
I like the idea of having a summary of code quality in the entire repo (mostly for "old" code). At least it might help more teams visualize their technical debt, monitor it and fix it!
I think this is interesting but still in its early stages.
Copilot Upgrades
Copilot also got some upgrades that are worth talking about. The most boring for me is "Plan Mode". Why? It's not new, we have this in Claude code, Cursor and Copilot with a custom chatmode that I have used a lot.
Copilot Code Review has improved, and I can already tell in my recent PRs. You can see the session now, and it runs CodeQL to find quality issues, and run other linters (e.g. ESlint). If you didn't know, GitHub Advanced Security is a paid feature and for private repos, it was the only way to have CodeQL and run it for code scanning and security analysis. It's really cool we get that with Copilot now, at least a piece of it. Code review now takes longer too since it's running these tools, but it's better IMO.
They also introduced Copilot custom agents, I mean... nothing new as well 😅. Sure it's new for Copilot, but we had sub-agents with Claude code, and agents in Cursor, Codex or others. But yeah, it's cool to now have custom agents in .github/agents. From what I have tested, they are basic for now, not a lot of configuration like configuring the model for Copilot coding agent (for VS Code you can), besides setting up more MCP servers.
We also have Copilot metrics! It's in public preview for all paid plans it seems, but I haven't looked at it.
announcement from their slides at GH Universe
I think it's cool to see how our teams and individuals are using AI. Still, it probably lacks many other useful metrics, but oh well. It's a good addition.
Conclusion
So yes, there are tons of new cool features and stuff. If you're thinking about adopting AI coding tools at your organization, give Copilot a try. All these new features, use and experiment them all. I'll continue to use copilot/AI on multiple tasks and experiment with custom agents and the rest. It already augments myself and my team, so incorporating more agents in our software development lifecycle could be even more beneficial.
Last but not least, there was a cool talk from GitHub Next (their R&D) team about Continuous AI 🙂. Cool ideas and prototypes overall, they are working on things I'll keep my eye on. Nothing I'll experiment soon though.
Top comments (0)