The pull request is the governance gate now, whether or not issue tracking is dead

CodeRabbit, an AI code review vendor, announced a control layer it calls Agentic Change Management on Wednesday, pitched as a way for teams to understand, govern and ship software written by human developers and agents alike ^[3][1][2]. The product matters less than the framing bolted to it: CEO Harjot Gill told The New Stack that "issue tracking is dead and agents are building software faster than teams can align on what should be built," which leaves the pull request as "the last real chokepoint in the SDLC" ^[5].

State the incentive plainly. CodeRabbit sells review, and its argument is that review is the one place left where control can still be applied ^[1]. The supporting reasoning, per Gill, is that as the marginal cost of producing code approaches zero the traditional SDLC breaks down, because it assumed code was scarce and expensive ^[6]; when code is abundant, human judgment becomes the scarce resource and should be spent on intent, architecture, behaviour and execution trade-offs rather than line-by-line reading ^[7]. The company also argues that trackers such as Jira, GitHub Issues and Linear were built for a simpler time, when planners had room to curate a backlog before assigning work ^[4].

You do not have to accept the word "dead" to test the mechanism, because it is countable in any repo. Take one quarter of merges on your default branch and count three things: pull requests opened with no linked issue, pull requests where the linked issue was created after the first commit, and the gap between a pull request opening and the first human comment that argues about scope rather than syntax. If code routinely arrives before anyone wrote down why it should exist, the pull request is already your only governance gate, and the vendor is describing your repository rather than selling you a future.

What the product does with that gate is more interesting than the slogan. CodeRabbit says the layer validates changes from developers and agents using repository-wide context, organisational standards, pre-merge checks, team knowledge and evidence from isolated test environments ^[9], that coding agents can clear findings through automated fix-and-re-review loops before a human looks ^[10], and that it rests on proprietary codegraph technology tuned over years to cut codebases into trust boundaries ^[12]. Gill describes adversarial review agents working with coding agents in a loop to automate review, validation and remediation, with humans moved to higher-order judgement of intent, behaviour, risk and outcomes ^[11]. Read the ordering carefully: the human no longer sees the agent's first attempt, only the diff that survived another agent's objections ^[2]. That is less review surface, and also less evidence about how the change was reached.

The company calls the pull request an "auditable decision point" where teams settle the quality bar, the risk carried, whether the change deserves human attention and whether it ships ^[8]. Audit is a records claim, so watch whether the record exists: approver identity, the standard applied, the test evidence attached, and, crucially, what was rejected and never merged. Watch also where change requests originate, since the source anticipates instructions arriving from non-technical staff, coding agents, trackers, support systems, telemetry pipelines and self-healing runtime logs ^[14]; each of those needs an owner when a merge goes wrong. And watch the trackers themselves. Teams under regulatory or contractual traceability duties will keep them regardless of who declares them finished, which turns the pull request into a second bookkeeping system rather than a replacement for the first. The pull request has carried mechanical weight since Torvalds wrote ./git-pull-script during Git's creation in 2005 ^[13]; making it the sole policy gate is a heavier load than it was designed for.