For as long as there’s been software, it’s been built for one kind of user: a person. That’s changing fast, but unevenly. Coding agents like Claude Code and Codex went from experiment to everyday tool in less than two years, while the systems underneath them have yet to change. So a lot of agent-scale work is now running on infrastructure that was sized for people.
Identity systems, for example, assume one login per employee, but one engineer might now run a dozen agents—or more—around the clock, each needing its own access. Developer tools assume a person writing and reviewing code at human speed. Nearly every layer of the stack has assumptions like these baked in, and agents break them. It is both necessary and exciting to rethink the entire development stack from the ground-up.
Developers are feeling this need very acutely. In May, Anthropic merged about 2,500 code changes a day into its monorepo. By September it was merging more than 30,000. In an engineering post this month, the company said the slow part is now everything around the code: source control, builds, CI. So it built its own replacement for Git. Very few companies have the means to build a Git replacement, and Anthropic’s isn’t generally available.
Everyone else is stuck with Git and GitHub as they are. Incident trackers counted 48 major GitHub outages in the year ending this April, about one a week. If GitHub is the heart of the software world, it’s showing signs of congestive heart failure: still pumping, but no longer able to keep up with what the body needs.
For the teams downstream, the problems pile up. Agents can open more pull requests in a day than a team can read in a week, so the review queue keeps growing and the human reviewer becomes the slowest part of the system. CI costs and wait times spike as each agent session kicks off its own builds. There was a time when a GitHub outage might cost a team an afternoon, but that loss is now multiplied by every agent that’s also left idle. And when an agent’s session ends, the context goes with it, so the reviewer is left squinting at a diff, trying to work out what the agent was asked to do and why it did it that way.
Source control is oft-discussed because developers are the first group to face these challenges, but data infrastructure is right behind. Databricks and platforms like it were built for data teams running scheduled pipelines and dashboards on shared clusters. An agent needs something different: an easy-to-use, ultra-fast data stack of its own for each task, that operates seamlessly across distributed data, and doesn’t become prohibitively expensive as data volumes scale. That’s what Spice AI, another Madrona portfolio company, is building.
Shifts like this tend to favor whoever starts fresh; both in terms of product and also overall vision. An incumbent has to fix its architecture one outage at a time without breaking anything for millions of existing users. A new company can design for agent traffic from day one. That’s what Thomas Dohmke set out to do when he left GitHub to start Entire, and with this week’s release, Entire is a full GitHub competitor.
Until now, Entire worked alongside GitHub. Its first product, in February, was an open-source tool that saved an agent’s context next to your commits. In July it added a distributed network that could mirror your repos. With the October release, you don’t need GitHub at all. You can host, review, merge, and deploy code on Entire, a Git-compatible forge with its own backend, review process, and memory, designed on the assumption that agents will write most of the code while people stay in charge.
It helps that Thomas has been on the other side of this. He was GitHub’s CEO for nearly four years, grew it past 180 million developers, and turned Copilot into a multi-billion-dollar business. That last part comes with some irony. Copilot did as much as anything to start the agent coding wave, so Thomas watched this traffic build from the inside long before it showed up as outages.
Here’s what’s in today’s release:
Hosting runs on Entire’s own decentralized Git network, with repos copied across regions (the US, EU, Australia, and India are live now). If one region has a bad day, your team keeps working, and you choose where your code lives. It’s fast, too: on ForgeMark, Entire’s open-source benchmark, a single repo handled about 570,000 clones an hour and 586 pushes a second.
Trails replace the pull request, and they’re the part we’re most excited about. Anyone who has opened a giant agent-written PR with no idea what the agent was going for knows the problem. A Trail keeps the intent and outcome of the agent’s session next to the code, gives it a confidence score, and runs automated checks, so people can focus on the changes that need a human. A Trail can also span several repos, which matters because a single agent task often touches a service, its client library, and the docs at the same time.
Then there’s Continuous Memory. Right now, most of what an agent learns about your codebase disappears when the session closes, and the next session starts from zero. Entire Graph maps how the codebase fits together, and Entire Brain links that map to the reasoning behind each change. Entire calls it the next step after continuous integration and continuous deployment. That sounds like a big claim until you’ve watched an agent make the same mistake three sessions in a row.
An important discipline for a startup team is prioritization. Entire has proven to be very smart about what to build but also what not to build. Instead of launching its own CI, it works with Buildkite, Depot, and CircleCI, and it’s building a plugin marketplace that goes much deeper than logins and APIs. Switching forges is a big ask, and letting developers keep their pipelines makes it a lot easier. Developer experience is Entire’s north star.
Amid all these transformative developments it’s easy to forget how young this company is. Entire launched in February with the largest seed round in developer tools history and one open-source tool. By July it had a new Git backend in public preview. Now, eight months after launch, it’s a full forge, and developers who started by mirroring a single repo are now telling the team they want to move everything over.
We’ve known Thomas for more than a decade, since Microsoft acquired his first company, HockeyApp. When he started Entire, Madrona was the first investor to commit. Even with that history, the team’s speed has surprised us. Thomas and a group of engineers mostly from GitHub and Atlassian have shipped three major releases in eight months, each a bigger swing than the last. For a long time GitHub seemed to be an unquestioned cornerstone of the developer’s workflow and many have used it for most of their careers. But as agents have transformed their workflow, Thomas saw an acute need and opportunity to serve them in a new way.
It took GitHub about a decade to become the default home for the world’s code. But the agentic era has been defined by moving at warp speed. Engineering teams are choosing their next forge right now, plenty of them in the middle of an outage, and we think a lot of them will choose Entire. You can see the October release at entire.io.
Related Insights
