New: Impact shows what your team fixed after Optibot's reviews.
All changelog entries
Introducing Impact
New Release Impact Insights Conversations Review Memory

Introducing Impact

A new Impact page shows what your team fixed after Optibot's reviews, and review threads now get a reply only when you ask for one.

Impact is a new dashboard page that shows what your team fixed after Optibot's reviews, where risk is building, and how AI-written code compares. Review threads are quieter too: Optibot answers when you ask, and acknowledges the rest with a 👍.

New

Impact

Open Impact under Workspace in the sidebar. Each section leads with a plain-language headline, then the chart behind it.

The Impact page in the Optibot dashboard. The headline reads: Your team fixed 212 issues Optibot caught before merge, 31 of them blocking. Beside a daily chart of blocking and non-blocking fixes, three numbers: 67% of findings fixed, 31 blocking fixes and 96 PRs with findings, each compared with the previous 30 days.
Impact opens with what your team fixed in the period you pick. Sample data.
  • What your team fixes. Fixes made after Optibot's reviews, broken down by kind, such as bugs, security issues and missing tests, with the kind that grew the most.
  • Where risk is building. The folders with the most blocking fixes, and the ones where blocking fixes are rising.
  • AI-written code. On GitHub, Impact recognizes pull requests written by coding agents such as Claude Code and Cursor, and compares how many fixes their code needs with code that has no AI signal.
Two Impact sections side by side. What your team fixes: a waffle chart and table of fixes by kind, led by duplicated code at 31%, then bugs, tech debt and security issues, with an Add a guideline link. Where risk is building: a before and now chart of blocking fixes by folder, with api/billing rising from 3 to 11, and an Add directory rules link.
Each section ends with a next step, such as a guideline or directory rules for a rising folder.
The AI-written code section. Its headline reads: Cursor agent code needs the most fixes, about 1.5x as many fixes per changed line as code with no AI signal. Rows for Cursor agent at 1.5x, Claude Code at 1.1x and Codex at 0.8x are plotted against a 1x line for code with no AI signal. Codex is marked Few PRs.
How code from each coding agent compares with code that has no AI signal.

Review threads that respect your time

Optibot now replies in a review thread when you expect an answer: a question, a point you want it to weigh in on, or a request to explain or check something.

  • A 👍 for updates. Reply to a finding with thanks, "fixed", or your decision, and Optibot reacts with a 👍 instead of posting a reply.
  • Deferrals stick. Defer a finding to a ticket, a TODO in the code, or a follow-up PR, and the next review treats it as settled.
  • Only when you mean Optibot. Teammates talking to each other, and commands meant for other tools, no longer get a response.
Three review threads on Optibot findings. In the first, a developer asks whether a bug also affects annual plans, and Optibot answers. In the second, a developer replies Fixed, thanks, and Optibot reacts with a thumbs-up and posts no reply. In the third, a developer defers a security finding to a ticket and adds a TODO; Optibot reacts with a thumbs-up and the finding is settled on the next review.
A question gets an answer. An update or a deferral gets a 👍. Illustration.

Improvements

  • Review memory that keeps learning. Review memory keeps picking up new lessons on your busiest repositories, and new repositories start out with more of your team's history.
  • More reliable review posting. A temporary GitHub error while posting one review no longer holds up later reviews on the same pull request.
  • Change plans during a trial. Choosing a higher tier or annual billing during your trial starts that plan right away, and checkout shows the amount charged.
Previous A New Dashboard and Shorter PR Summaries

See Optibot in action

Get a personalized demo and see how Optimal AI fits your team.

Get a demo