GitButler said on Thursday that recordings of all 10 main talks from Git Merge 2026, the version control conference it hosted with OpenAI last week in Lisbon, are now available on its YouTube channel.
The event drew sponsorship from GitHub, GitLab, Tangled, Cursor, Pierre and Google, Scott Chacon, a co-founder of GitHub and GitButler, wrote in a blog post summarizing each talk. Talks from JJ Con, a day of sessions dedicated to the Jujutsu version control system held before the main conference, are also online and summarized in a separate post.
A thread running through several sessions was what happens to Git as software agents write more of the code. Tomas Reimers traced code review from printed Fagan inspections through emailed patches to web interfaces, and argued that review does three jobs: checking correctness, keeping changes manageable and spreading understanding. With agents producing more code, he called for giving them direct access to forges, for tools that guide reviewers’ attention and for reviewing screenshots or running software alongside diffs.
The programmer brian m. carlson laid out Git’s plans to adopt Rust and the portability problems for platforms without Rust support. He covered sharing errors between C and Rust and rewriting object conversion in Rust for SHA-256 interoperability, with future work on safer index and pack parsing and a tool to migrate repositories between hash algorithms.
Emily Shaffer of Google explained why moving to SHA-256 means converting repository history and accounting for submodules, old commit references and tools that assume SHA-1 identifiers. She described Google’s temporary use of Copybara to bridge incompatible repositories and outlined a gradual migration, with servers storing the source of truth in SHA-256 while clients and vendors switch at their own pace.
Mr. Chacon used his own talk to introduce git-meta, a proposal with a Rust reference implementation for attaching namespaced metadata to commits, branches and paths. The closing demo used imported pull-request data to blame lines by pull request rather than by commit.
Elijah Newren showed ways Git wastes work in large repositories, from merge-base searches that walk unnecessary history to a shallow-clone push problem that can download gigabytes only to send them back to the same server, and presented fixes. Patrick Steinhardt described GitLab’s work to make Git’s reference and object storage replaceable, demonstrating a deliberately basic SQLite backend and a larger setup that lets multiple servers share repositories held in object storage.
Other speakers included Christian Schilling on Josh, which carves self-contained filtered histories that can be merged back into the original repository; Raphaël Gomès and Pierre-Yves David on how Mercurial shares rewritten history across clones; Anirudh Oppiliappan on separating hosting, CI and interface at Tangled through open APIs and AT Protocol; and Toon Claes, who proposed new git log options for filtering to one’s own branches and collapsing unrelated history.
Mr. Chacon thanked the speakers, sponsors and attendees. “We’ll see you at the next one,” he wrote.