What happened
Cognition announced Devin Review 2.0 on November 5, 2026, and says it is now the default tool for PR review when working with Devin. The company frames the change around a growing mismatch: PR throughput is increasing while human review capacity is not. Its proposed response is to use an automated reviewer to find bugs and provide a guided tour of a PR, so a human can review with more context rather than skip review or inspect fewer changes.
The headline change is cost. Cognition says the cost of reviewing a PR has fallen 82% since August, to a typical $0.25–$0.75. It attributes the reduction to a better mix of models and harness optimizations, and says quality did not suffer. In its reported averages, bugs found per PR rose from 1.22 to 1.32 after the cost reductions. The company also reports that about 80% of PRs cost less than a dollar to review, counting all pushes, and that median latency is now 94 seconds, down from 139 seconds in August.
Teams can start a review with a /devin review comment in GitHub or GitLab, or from the Devin Review page in Devin Cloud or Devin Desktop. They can also configure automatic review when a PR is ready, or on every push. Rules can include draft PRs or filter by file paths, branches, labels, and other criteria; Cognition says customers use these filters to exclude low-value PRs and manage spend. When a review runs again on the same PR, it now checks only the changed parts by default. Cognition says this reduces noise and cost while focusing findings on the latest changes.
The release adds controls for different risk and review needs. A configurable per-PR spend limit is intended to prevent an individual PR from running up costs. Devin Review Ultra is a higher-cost, more thorough effort level for areas such as billing or security-sensitive code. Custom Reviewers, now in beta, let teams define specialized checks using their own instructions, with separate automatic-review rules. Cognition says these findings appear in Devin Review results and can feed Devin’s automatic fixing. Used together, Devin can investigate a finding and decide whether to fix it, dismiss it, or escalate it to a person.
What the archive adds
This is a new product update within a recurring Cognition thread, not a restatement of the earlier coverage. Cognition’s Pylon integration connected support tickets to Devin coding sessions; its efficiency-based auto-approvals tied spend controls to estimated output. Devin Review 2.0 applies cost controls to another part of the workflow: automated review of pull requests.
Trend signal
The update supports agent economics moving to completed work: Cognition emphasizes cost per reviewed PR, not cost per model call, and says incremental checks make repeated review more affordable. It also touches verification bandwidth, since automated review is positioned as a way to expand review capacity. The release does not establish that the system catches enough consequential defects to replace human verification.
Why it matters
For teams already using Devin, the reported price range makes it possible to test whether automated review can cover more of the PR lifecycle without an open-ended bill. Incremental review and filters matter as much as the headline discount: teams can choose which changes to inspect and avoid paying to reprocess an unchanged PR. A per-PR cap gives an additional guardrail.
The product also separates routine review from higher-risk checks. Ultra offers a more thorough pass at higher cost, while Custom Reviewers lets teams encode local expectations—such as safe migrations or accessibility checks—rather than relying on a generic bug-finding pass. That could help teams turn repeatable review practices into automated checks. The practical question is whether teams can validate the findings and tune the rules without adding more review overhead than the system removes.
Limits and uncertainty
The cost, latency, and bug-count figures come from Cognition; the supplied announcement does not provide an independent evaluation, sample size, distribution by repository or language, or a definition of a “bug.” An increase from 1.22 to 1.32 average findings per PR does not by itself show that the additional findings are correct or important. The post also does not quantify false positives, missed defects, or how often humans accept Devin’s fixes. Custom Reviewers are in beta, and the announcement gives no pricing schedule beyond the typical review range and the statement that Ultra costs more.
Skill to build now
Choose a small set of representative PRs and define what counts as a useful finding before enabling automatic review broadly. Compare Devin Review’s findings with human review, track accepted and dismissed findings, and set a per-PR spend cap. Start with a narrow file-path or branch rule; use the results to decide whether review on every push is worth the extra coverage and cost.
Reader takeaway
Devin Review 2.0 turns Cognition’s lower per-PR cost into a broader automation proposition: review can run by default, incrementally, and with configurable limits. The useful test for a team is not the 82% reduction alone, but whether the findings are reliable enough to improve review coverage at a predictable cost. Treat the announced metrics as claims to validate against your own code and review process.