This trip almost didn't happen.
A few weeks before I was due to fly from Bangalore to California, I got an email saying the Grace Hopper Celebration had been cancelled. My bookings were made. My bags were mentally packed.
So a group of us from Slack did what engineers do best. We found a workaround, and the workaround was GitHub Universe.
I'm so glad we did.
The opening keynote, and what it told me about the future of tech
The first thing I remember is the line for the keynote. It was very long. The second thing I remember is how quickly the conversation turned to one idea, and I heard versions of it all week.
Writing code is cheap now. Deciding what to build, and how to build it, is the real work.
That shifts our priorities. Knowing the depth of one codebase matters less than being able to work across its breadth: talking to real customers, and solving problems for real people.
GitHub framed the whole event around an GitHub's term for treating delivery as one system where agents plan, write, review and verify work alongside people, measured by finished outcomes rather than lines of code.. Their message was that what matters is shipping completed work, like a feature or a fix, not producing more code.
That sounds simpler than it is. If code is no longer the bottleneck, then planning, alignment, review and verification are.
The session that left the biggest mark on me
On day one I found my way to a breakout session by Maggie Appleton from GitHub's research and development team, which prototypes what software development could look like next.. It was held in a silent-disco space, so we all sat there with headphones on, completely absorbed.
She showed us Chopin: a real-time, multiplayer planning environment where coworkers and agents write plans together. GitHub Next has open-sourced it.
The problem with planning alone with an agent
When you plan with an agent today, it's usually just you and the model. The agent asks you a long list of questions. Then it hands you a huge wall of text and asks you to approve it, yes or no.
Your teammates can't see any of it. And nothing keeps that plan up to date as the design or the architecture changes.
Maggie put a number on why this feels so draining.
No wonder agentic work feels exhausting some days.
What Chopin does differently
That last one made me smile. We spend so much energy making agents faster. Here was a team making an agent more polite.
Later I got to have a one-on-one with Maggie. She was generous with her time, and honest that Chopin is still a research project, not a product. The value isn't the tool itself. It's the question it asks:
What would happen if we treated planning as a shared workspace for the whole team, agents included?
I think that question could change how we imagine teamwork.
GitHub NextThe research team behind Chopin, prototyping what software development could look like next.githubnext.comVerification is the new craft
The second theme I kept hearing was verification. As code gets easier to generate, knowing it actually works becomes the real job.
- Cross-model review. Let models from different families review each other's code, because models from the same family tend to share the same blind spots.
- Agentic testing. The agent explores the product, finds issues, and those findings become Tests that lock in behavior that already works, so a later change that breaks it fails the build..
Microsoft Research went a step further. They showed PDD Kit, which turns a developer's intent into a A precise, machine-checkable statement of what the code must do. A prover can then check the code against it, instead of a person reading every line. and then proves that the generated code meets it. It only works for Rust today, through the Verus verifier, and it's early.
But the idea behind it is powerful. You agree on the contract first, then you review the contract instead of every line of code.
Microsoft's PDD Kit Makes Formal Math Proofs a Gate for AI Rust CodeMicrosoft Research unveils PDD Kit, bringing formal verification to AI-generated code via Verus for Rust.AlphaSignal
More code is not more value
One data point from a session by Jellyfish stopped me in my tracks.
More activity does not automatically mean more impact.
The same talk shared that most teams believe AI improves delivery, but far fewer actually measure it. The suggestion was to measure in this order:
- Adoption first,
- then speed and quality,
- then return on investment.
GitHub's answer to the review side of this is Splitting one large change into a chain of small pull requests, each built on the one before it, so every piece can be reviewed and merged on its own..
Things close to my heart
I brought home a lot of notes, a few new friends, and a GitHub keyboard kit.
I assembled it this weekend. It took me close to two hours. Some of the keycaps didn't fit, so I used the spare set they provided. It's wired only, no Bluetooth. But it works, and I built it myself.
That feels like the right metaphor for this moment in our industry. The pieces don't always fit on the first try. It takes longer than you expect. But you figure it out, especially when you're not doing it alone.