I had the chance to attend GitHub Universe in San Francisco, and wow – what a week. Between the keynote energy, hands-on demos, and meeting the people behind the tools we use every day.
In this post, I’ll take you through some of the most exciting highlights, share my take on what they mean for developers and teams, and explore how AI-driven workflows are reshaping how we build software today.
A Universe expanding faster than ever
Let’s talk numbers, because they are wild!
From the latest Octoverse 2025 Report:
- 630 million total projects on GitHub
- 4.3 million AI projects (yep, everyone’s experimenting)
- 43.2 million pull requests merged every month
- And my favourite stat – a new developer joins GitHub every second
Even crazier? 80% of new developers use Copilot in their first week. AI isn’t just “part of” development anymore – it’s baked right in.
The lines between code, creativity, and AI are blurring fast… and honestly, it’s kind of beautiful.
“Humans Drive, Agents Deliver” – Welcome to Agent HQ
If 2024 was about Copilot becoming your coding partner, 2025 is the year of agents – your extended, specialised team members.
One of GitHub’s big drop this year was Agent HQ, and it’s a complete rethink of how we work with AI.
And Agent HQ isn’t just a GitHub exclusive club. Anthropic, OpenAI, Google etc – every big-name AI player, cosied up in one ecosystem, making it possible to govern and orchestrate agents however your workflow demands
And here’s what’s awesome – it’s an open ecosystem. They’re not locking you in.
They’re giving you options and letting you pick what works best for your team. Pretty awesome, right?
Mission Control – Your AI Command Center
At the heart of Agent HQ is Mission Control.
It’s like the cockpit for your agentic workflow – following you across GitHub’s web UI, VS Code, mobile, and even the CLI.
From Mission Control, you can:
- Assign work to multiple agents simultaneously
- Track progress and manage permissions
- Enforce compliance and governance without context switching
Think of Mission Control as your single dashboard for the agentic circus. Assign tasks to agents, track what they’re up to, and steer the ship.
Does that make sense? Instead of switching contexts constantly, you’ve got one unified view of all your AI assistants.
Imagine one dashboard to rule your entire AI squad. Sound like a dream? It’s not. It’s here.
Custom Agents: Because One Size Doesn’t Fit All
This feature got me particularly excited. You can now create custom agents with specific prompts, tools, and context for your team’s unique needs
Imagine this: You’ve got coding standards, compliance requirements, maybe some quirky conventions that make sense for your organisation (we all have them, don’t lie).
Instead of repeatedly telling every AI agent about these, you define it once in a markdown file. Done.
These custom agents can:
- Use organisation-specific tools
- Connect to specific MCP servers
- Enforce your coding conventions automatically
- Act like specialised teammates
No complex setup. Just drop a config file in your repo or organisation settings. That’s it.
Smarter Code Reviews
The new Copilot Code Review changes are great, blending LLM intelligence with deterministic tools like ESLint and CodeQL.
It’s like having a reviewer who understands both “this code is technically correct” AND “yeah, but this is going to be a maintenance nightmare in six months.”
The review happens before your human teammates even see the PR. Agents do the first-line review, catch the obvious stuff, and then hand it off to humans for the judgment calls that actually need human judgment.
Smart, right?
Plan Mode: Think Before You Code
Ever jumped straight into coding and realised halfway through that you misunderstood the requirements? Yeah, me too. More times than I’d like to admit.
Plan mode addresses this. You co-create a step-by-step plan with Copilot *before* writing any code.
Catch the gaps early. Approve the plan. Then implement it locally or let a cloud agent handle it.
It’s the software equivalent of “measure twice, cut once.” Except you’re planning first, then coding.
Agentic Workflows – Natural Language Did What?
If you’ve read my recent post on Agentic Workflows, you know how much I love this concept.
You can literally tell GitHub, in plain English, what you need done — and it orchestrates the agents to make it happen.
“Deploy that new service.”
“Run the tests and open an issue if anything fails.”
“Validate our Terraform modules.”
It’s natural language as the new CLI. Wild, right?
Confidence & Control
If you’re leading a team (or dealing with folks who worry about governance), GitHub rolled out some serious enterprise features:
GitHub Code Quality
Organisation-wide visibility into code quality, maintainability, reliability, and test coverage. Finally, metrics that actually matter.
Copilot Metrics Dashboard
Track adoption and impact across your org. See what’s working, what’s not, and prove ROI to whoever needs convincing.
The Agent Control Plane
This is huge. Enterprise AI Controls let you:
- Set security policies
- Enable audit logging
- Manage which models and agents are allowed
- Control data access
And crucially? You do all this without slowing down your developers. Because nothing kills productivity faster than security theater.
The Vibe: If You Can Dream It, You Can Build It
The quotes they kept repeating throughout the event stuck with me:
“Humans drive, agents deliver.”
“The best developers never stop experimenting.”
“AI doesn’t write the story. You do.”
That last one? That’s the whole point, really.
We’re not being replaced by AI. We’re being *amplified* by it. The agents handle the repetitive stuff, the boilerplate, the “I’ve written this same code pattern a hundred times” work.
We focus on the creative stuff. The problem-solving. The “what should we build next?” questions.
What This Actually Means for You
So, what does all this mean for your day-to-day work?
If you’re a developer:
- Less time fighting with boilerplate code
- More time solving interesting problems
- AI agents that understand your team’s specific needs
- Unified interface for all your AI tools
If you’re leading a team:
- Visibility into how AI is being used
- Governance controls that don’t block productivity
- Metrics to prove value
- Security and compliance baked in
If you’re at an enterprise:
- Path to consolidate AI tools without forcing elimination
- Reduced vendor lock-in risk
- Test agents without committing to one forever
My Perspective
GitHub Universe 2025 wasn’t just about flashy demos (though those were fun). It was about GitHub acknowledging reality: AI-enhanced development is here to stay, and we need better tools to manage it.
Agent HQ feels less like “here’s our AI agent” and more like “here’s how to make all AI agents work together without losing your mind.”
GitHub isn’t just sprinkling AI features into existing tools – it’s building what feels like the operating system for AI‑driven development. Agent HQ addresses a real pain point: too many AI tools, too many disconnected workflows.
The multi‑agent model is smart, the governance features are ready for enterprise use, and the open design means you can choose the agents that fit your needs.
1 thought on “GitHub Universe 2025: A Front-Row Seat to the Future of Coding”