What's next?
TL;DR
Theo thinks GitHub is still the best Gen-2 source-control product — even while it’s breaking — he says GitHub’s outages, random merge reverts, and agent-driven load are making people leave, but GitLab and Bitbucket mostly fail as true drop-in replacements.
GitLab gets torched on UX, not viability — Theo calls it “a bicycle” everyone says is sensible but hates using, then walks through examples like hidden loading layers, impossible commit-history browsing, and a 12.8 million-line codebase with 528,000 commits that feels too old and massive to fix quickly.
Bitbucket is framed as an Atlassian bundle discount, not a product win — Atlassian’s own pitch leans on Jira integration and cost savings, including a comparison that stacks GitHub Enterprise Cloud with extra add-ons to make Bitbucket look 10x cheaper.
Forgejo + Codeberg is the first alternative that genuinely surprised him — after expecting to dislike it, Theo found the Go-based stack fast enough, much smaller than GitLab, better on core release navigation, compatible with GitHub Actions-style YAML, and impressive enough that he donated $1,200 plus $400/month.
The real next wave may not be a GitHub clone at all — Theo argues companies like Pierre, Graphite, and Entire are building Gen-3 primitives for agent-heavy development, from Pierre’s code.sto handling 9 million repos in 30 days and 15,000 repos/minute to Entire’s attempt to preserve agent reasoning alongside code.
The hardest thing to replace isn’t git hosting — it’s GitHub’s social graph — Theo ends on the loss of the shared home where repos, issues, identities, and contribution history all lived together, saying the open-source world is entering a painful “graduation moment” of fragmentation.
Summary
Why people are suddenly serious about leaving GitHub
Theo opens with the mood shift: GitHub no longer feels like the safe default when merges can get reverted and outages last days instead of minutes. Before comparing options, he defines the actual bar: hosted git, PR workflows, community features like profiles and stars, and ideally CI/CD, stability, openness, and some kind of AI-native future.
GitLab: the sensible alternative that feels awful in practice
GitLab gets the “bicycle” analogy: everyone agrees it’s a valid option, but the people who actually use it make it sound miserable. Theo leans on examples from Josh and Jason Cox, showing hidden loading layers, confusing navigation, infinite scroll for commit history, terrible release pages, and UI weightings so flat you can’t quickly find the things that matter.
The deeper GitLab problem is generational, not cosmetic
Once he clones the GitLab repo and sees 7,59,000 git objects, 12,78,69 lines of code, 3.88 million lines of Ruby, 1.16 million lines of JavaScript, and even Vue 2, the point lands: this is an old, huge system. Theo argues GitLab is like Azure to GitHub’s AWS — similar, maybe a bit better on uptime, but otherwise just a worse product from the same generation.
Bitbucket: more Jira strategy than source-control vision
Theo barely has to stretch here because Atlassian’s own messaging does the work: “Git solutions for teams using Jira.” He mocks the pricing math, notes the “10x savings” comparison is padded with GitHub add-ons, and concludes Bitbucket’s real value is only for companies already deep in the Atlassian stack.
Forgejo and Codeberg unexpectedly become the heroes
This is where the video turns. Theo expected Forgejo/Codeberg to be worthy but rough; instead, after making an account and poking around, he finds a smaller, cleaner Go-based codebase, a release view he actually likes, good-enough code review, GitHub Actions-compatible YAML, and nonprofit governance that feels trustworthy after Gitea’s “rug pull.”
From skepticism to wallet-open support
Theo gets genuinely excited here: he praises Codeberg’s transparency, self-hosting story, and better uptime than GitHub, then donates $1,200 and commits $400/month. His takeaway is blunt: if you want an open, self-hostable Gen-2 option, don’t touch the “pile of Ruby slop” in GitLab — use Forgejo.
Pierre, Graphite, and Entire are building the Gen-3 future
Theo shifts from “what can replace GitHub today?” to “what comes next?” Pierre’s code.sto is his favorite example: 9 million repos in 30 days, peaks of 15,000 repos per minute, and a model built for agent throughput rather than human-era git hosting. Graphite, now inside Cursor, and Entire, founded by former GitHub CEO Thomas Dohmke with a $60 million seed, are both presented as bets on workflows where agents, context, and code history get rethought from first principles.
The real loss is the community layer, not the tooling layer
Theo ends on the emotional core: GitHub wasn’t just a remote, it was the shared home for open-source identity. Once projects scatter across Forgejo, GitLab, self-hosted instances, and whatever comes next, you lose the simple magic of clicking one profile and seeing a developer’s whole body of work — and he’s clear that no aggregator or clever tool will fully restore that.
Was This Useful?
Share
Keep Reading
Make Alcreon Yours
Tune your feedFive quick questions, and the feed ranks what matters to you first.Or just get notified
The weekly Echo. Signal worth keeping in your inbox.
Every new piece, announced on X.
Read Next
See all
Playbook
The Retirement Email Isn't a Warning
Model retirements now arrive every few weeks; the config-eval-rehearsal loop turns each deprecation email from a fire drill into an afternoon swap.
Playbook
The Cheapest Model That Passes
OpenRouter lists 400 models behind one API. The fix for choosing isn't a better leaderboard, it's a four-step protocol that ends in a real eval.
Playbook
Cheap Models, Hard Tasks
Most agent workflows route every step to the frontier model by default. The bill scales with how chatty the agent gets, even when most steps don't need that brain.