GitHub using lessons from GitHub to build a GitHub that competes with GitHub
Former GitHub CEO and CTO are cooking
John and I kicked off "Open Source Ready Live" - a more freeform companion to our podcast where we dig into what's happening in open source and AI. The plan is to host it live on YouTube, and potentially expand to other channels.
It’s Not About the Product, It’s About the Models
Jason Warner (former GitHub CTO, now CEO of Poolside) dropped a thought-provoking thread about the AI infrastructure landscape, and Cursor became the focal point of the conversation. But reiterating from what was said on stream - Jason wasn’t calling out Cursor specifically. He was highlighting a broader industry shift.
Remember when everyone said “the model doesn’t matter, the product is where the money is”? That thesis is being tested right now. Cursor raised $2.3 billion at a $29 billion valuation to train their own models - a massive bet that the infrastructure layer matters more than we initially thought.
The reality is we’re watching the AI infrastructure world play out in real-time. We can see where the holes are: in foundational model ownership, in infrastructure control, in the expensive game of training models. The companies that can navigate this - whether it’s Poolside, OpenAI’s upcoming IPO, or even Google’s TPU advantage - are the ones positioning themselves for the long game.
Jason’s point wasn’t about Cursor but rather “we’re all learning that this game is more complex than we thought.”
Entire: Competing with GitHub Using GitHub’s Playbook?
And speaking of complexity - let’s talk about Thomas Dohmke’s new venture, Entire.
Thomas, the previous CEO of GitHub, just announced Entire, a CLI tool that adds checkpoints and metadata to your git workflow. Think of it as version control that understands your AI coding sessions.
John made a great point during our stream: anyone who uses GitHub every day knows that the pull request model, while powerful, feels increasingly awkward in an agentic world. What if the next evolution isn’t just adding metadata to git commits, but rethinking the entire collaboration model for AI-native development?
Consider the parallels:
Git was version control for code
GitHub made git social and collaborative
Entire is adding AI-native concepts to version control
We’re potentially watching someone use lessons learned from overseeing GitHub to build something that could eventually compete with GitHub. And they’re doing it immediately after running GitHub. Wild.
The really fascinating bit? GitHub is now fully integrated into Microsoft’s Core AI division. They’re solving Microsoft problems, hitting Microsoft OKRs. Meanwhile, you’ve got former GitHub leadership (Jason at Poolside) and former GitHub leadership (Thomas with Entire) both making moves that suggest the next wave of developer tools won’t come from Redmond.
The Pattern We’re Seeing
There’s a broader pattern here worth noting. GitHub had early ownership of AI coding with Copilot. But they got lapped. Cursor came in with a better agentic experience. Then came the summer 2024 arms race - Code Rabbit, Graphite (acquired by Cursor), Continue, and a dozen others all rushing to hook into the PR workflow.
GitHub’s response? Agent HQ - essentially opening up GitHub as a platform for everyone else’s agents. It’s a defensive move disguised as openness.
And now we’re seeing the ecosystem attack from multiple angles:
Cursor bought Graphite to own more of the workflow
Thomas is building version control primitives that don’t need GitHub
The question isn’t whether GitHub will survive, but are they too big to fail? They remain the center of gravity for AI-native development to, or whether they become “just Dropbox for your code” as I mentioned in the stream.
Yet, the trend is growing that center of gravity is getting smaller.
What We’re Building
John and I are working a new project, Tapes (github.com/papercomputeco/tapes) - a telemetry system for agents to capture, search, and checkpoint agents sessions. If you’re interested in understanding what your agents are actually doing, or sharing context across your team with zero instrumentation, check it out. We’re at 100 stars and counting. We’d love your feedback.
Stay ready!