Summary of "What's next?"
Overview
The video argues that many developers are considering leaving GitHub due to reliability and operational issues—most notably:
- “Randomly reverting merges”
- Outages lasting days
It then presents a structured way to evaluate replacements, reviews several alternatives (highlighting what’s viable today vs. what’s still immature), and discusses a “next generation” shift in how source control may work with agentic AI.
What you’re looking for (criteria based on what GitHub does well)
The speaker starts by listing GitHub strengths that alternatives should match.
Core requirements
- Hosted Git with collaborative workflows
- shared remote
- PR workflow
- contribution + merging
- Community layer
- profiles/history/feeds/stars
- related social signals
- CI/CD
- often via GitHub Actions, though alternatives may differ
“Nice to have” requirements
- Stability (not down for long stretches)
- Open source / self-hosting option
- A vague “AI-native” angle (not fully fleshed out)
Sponsored/ads section (agent permissions topic)
A sponsor is used to highlight that agentic systems raise new questions, such as:
- How to verify an agent is the right actor
- How to ensure correct permissions when using MCP
- How to verify a user is actually the user
Sponsor mentioned: Work OS (login/admin onboarding).
Evaluation: GitLab (considered a poor “drop-in” replacement)
The speaker’s stance: GitLab has worse usability/UX than GitHub and is not a true replacement, even if uptime might be better.
Key technical/UX complaints (examples)
- Navigation/loading glitches
- content can disappear when using back/forward because requests don’t refetch correctly
- “Infinite scroll” problems
- commit browsing is slow and hard to locate historical items
- Project page readability issues
- key info (README/release info) is hard to find; often positioned too low
- Release view confusion
- unclear fields like “% complete”
- awkward/missing date placement
- release diffs described as unhelpful
- Commit/release navigation limitations
- missing “next commit” direction
- limited movement through the commit graph from release pages
- Slow performance / heavy legacy stack
- large codebase cited (hundreds of thousands of commits; millions of lines; Ruby/JS mentioned)
- cloning described as slow to start enumerating objects
- mentions Vue 2 and Ruby-based architecture
Conclusion
GitLab is characterized as “worse GitHub”—similar overall model, but worse UX and no compelling advantage besides possibly uptime.
Evaluation: Bitbucket (cheaper + Jira integration, not a true UX replacement)
The speaker summarizes Bitbucket Cloud primarily as a cost and Jira integration play.
What Bitbucket is claimed to provide
- CI/code savings (up to 10x) and lower per-user cost vs GitHub Enterprise Cloud
- DevSecOps tooling (secret scanning, dependency scanning, IaC scanning) without extra charges (per marketing)
Pricing critique
The speaker argues that the comparisons rely on assumptions and add-ons that inflate the “GitHub is expensive” framing.
- Example mentioned: 3500 build minutes/month per organization (with more purchasable)
Takeaway
Bitbucket’s value is mostly “if you’re already an Atlassian/Jira shop”; otherwise it doesn’t solve the “replacement” problem well.
Evaluation: Forgejo + Codeberg (recommended practical “open” GitHub-like alternative today)
The speaker strongly shifts preference toward Forgejo (software) and Codeberg (hosted service) as the most promising “Gen 2” open alternatives.
Forgejo/Codeberg highlights
- Forgejo self-hostable; Codeberg offers hosted Forgejo without setup
- Praised factors:
- Maintained and comparatively smaller codebase (example sizes cited like ~12MB server footprint; ~400k lines of Go)
- Go-based performance expectations (vs Ruby stacks)
- Release tab usability: “exactly the info I need” (timestamps, commits included)
- Search works; an RSS feed exists for releases
- Issues/code review UI feels like a GitHub clone, but “solid” and quick enough
- Some rough edges are mentioned (e.g., some slow parts, theme switching behavior), but overall:
- “I could see myself using this”
- “pick this over GitLab easily”
“Governance / origin” claims
- Forgejo is described as created after a governance shift (a for-profit takeover) and maintained under a nonprofit umbrella.
- The speaker emphasizes:
- freedom/free software approach
- faster security fixes
- stability work
- end-to-end testing
Actions/support and transparency
- Forgejo Actions use YAML similar to GitHub Actions.
- Codeberg hosted Actions are described as having partial downtime (for security/infra reasons) but better overall uptime than GitHub at the time.
- The speaker calls out status transparency as a strong point vs GitHub.
Practical recommendation
- The video explicitly encourages viewers: go try Forgejo and Codeberg, and even self-host if higher uptime is required.
- Actions compatibility is noted with common CI providers.
- The speaker suggests linking to third-party CI services (e.g., Depot/Blacksmith) via adjacent sponsor references.
Other alternatives briefly covered / dismissed
- Code Commit / Fabricator / Google Code: largely dead
- Pierre
- discussed as “paused” as a GitHub alternative project
- but positioned as important for building foundational Git/storage primitives for a future (Gen 3) source control model
- SourceHut
- not positioned as the right fit for the speaker’s needs
“Gen 2 vs Gen 3” framing (future direction)
A major theme is generational change in tooling UX and architecture, analogous to:
“Sublime → Atom → VS Code”
…and then a shift toward agent-first coding experiences.
Gen 2 (today’s GitHub-like model)
- Centralized repo hosting platforms that still follow classic Git workflows and PR/merge paradigms
- Examples treated as Gen 2: GitLab, Bitbucket, Forgejo/Codeberg
Gen 3 (rethink source control for agentic workflows)
The speaker argues:
- Git snapshots may lose “why” context when agents generate large outputs.
- Gen 3 needs durable context + different primitives for agent-driven coding.
Pierre / code.sto (foundational “Gen 3” building blocks)
Pierre is framed as building “primitives” for an ultra-low latency Git cloud, including:
- programmatic Git integration
- fast reads/writes
- concepts like ephemeral/in-memory writes, cold storage, and graph concepts
Throughput metrics are cited (millions of repos stored; repos/minute peaks) to claim it was designed for agentic scale, unlike GitHub’s layered approach.
Entire (agent context + durable history for “why”)
- Git preserves what changed, but not why
- Agents need shared durable context to avoid duplicated work
- The discussion references ecosystem ideas and comparisons to other attempts (e.g., Zed’s concepts like Delta DB / CRDTs and agent protocols)
Graphite (code review next-gen, evolving from GitHub)
- Graphite improved code review workflows significantly
- Initially built on GitHub APIs
- Added mirroring to reduce GitHub API latency
- Later acquired by Cursor
- Speaker speculation: Graphite/Cursor could become an alternative by building different “thinking,” not merely a better GitHub
The community cost (central emotional/analytical conclusion)
Even with technical alternatives, the speaker emphasizes a non-trivial downside:
- GitHub acts as a single social/community home for:
- profiles, histories, contributions, conversations
- issues and links between projects
- Leaving GitHub causes community fragmentation (“great fracturing”):
- projects split across platforms/instances
- manual import/aggregation won’t recreate the same network effects
- It breaks the “one username shows everything” model and makes it harder to verify legitimacy from a single profile.
Main speakers/sources (as stated in subtitles)
- Mitchell (creator of Ghosty)
- Josh (mentioned as a person who debates/supported the GitLab critique)
- Jason Cox (cited as someone who broke down additional GitLab problems)
- The “Atlassian” team / Atlassian sources (for Bitbucket positioning vs GitHub)
- Forgejo/Codeberg (key referenced projects; “Codeberg EV” nonprofit umbrella)
- Pierre / code.sto, Entire, Graphite, Cursor, Zed (referenced as product sources)
Category
Technology
Share this summary
Is the summary off?
If you think the summary is inaccurate, you can reprocess it with the latest model.