Gerrit is a code review system built on Git, and its central idea is different from every pull-request platform. In Gerrit the unit of review is a change — a single commit, identified by a Change-Id in its commit message, that is pushed to the magic ref refs/for/<branch> rather than to the branch itself. Reviewers score it, CI verifies it, and only when the submit requirements are satisfied does Gerrit merge it. Nothing reaches the target branch without passing the gate.
That model produces properties that branch-and-merge workflows do not. Because a change is one commit that is amended rather than a branch that accumulates commits, review history is a series of patch sets against the same Change-Id, and reviewers can diff patch set 3 against patch set 5 to see exactly what an author changed in response to a comment. Because merging is performed by the server under submit rules, the branch state is a consequence of policy rather than of who clicked merge.
Gerrit is also a Git server with a fine-grained permission system. Access is defined in project.config, projects inherit from a parent — ultimately All-Projects — and rights can be granted per reference pattern, so who may push, submit, force-update or bypass review is expressed precisely. Review data lives in NoteDb inside the repository itself, and the replication plugin mirrors repositories to other sites. It is the tool organisations reach for when review must be mandatory and provable, which is why it is common in telecoms, automotive, semiconductor and large open-source projects.
Why this skill matters now
The industry standardised on pull requests, and for most teams that is the right default. But a category of engineering organisations cannot use a workflow where review is advisory and merge is a button — they need every change gated, every approval attributable, and every branch state explainable to an auditor. That is Gerrit's territory, and it has not been displaced.
The practical demand comes from two directions. Platform and tools teams inherit a Gerrit instance that someone configured years ago, and now have to administer it: permissions in project.config, submit requirements, the Jenkins verification loop, replication, upgrades and reindexing. And engineers joining automotive, telecom, embedded or chip-design programmes hit the change-based model for the first time and are unproductive for weeks, because it does not respond to the habits they built on GitHub.
The gap is specific and teachable. Understanding Change-Id and the commit-msg hook, amending rather than adding commits, relation chains and rebasing dependent changes, and how submit rules decide what is mergeable turns a frustrating tool into a fast one — usually within a day.