PR Lens vs Graphite

Graphite and PR Lens are not competing for the same slot, and the confusion is worth clearing up because both get filed under "AI code review" in every round-up. Graphite is a stacking workflow with a merge queue and a reviewer attached. Its answer to a big pull request is to make it several small ones. PR Lens draws whatever pull request arrives, at whatever size it arrives. LinearB's 2026 benchmarks, from 8.1 million pull requests across more than 4,800 organisations, found AI-assisted pull requests running 408 lines at the 75th percentile against 157 for unassisted ones. Graphite attacks that number. PR Lens attacks what the number costs you.

What Graphite does well

Stacking, mainly, and it has been the best tool for it for years. You create a series of dependent pull requests, each building on the last, so you can keep working on the third while the first two are still in review. When an earlier one merges, the ones above it rebase automatically. That is a genuinely different way to work and people who adopt it rarely go back.

The merge queue is the other half. Graphite's docs describe it as automating the rebase during merge to prevent semantic conflicts and keep trunk green, and the interesting part is that it is stack aware: stacked pull requests queued together can be validated in parallel and fast-forward merged, because CI already ran against that exact changeset. Most merge queues are strictly sequential. This one is not, and on a busy trunk that difference is measured in hours.

Around those two things sit a CLI, a VS Code extension, an MCP server, an inbox that tells you what is waiting on you, and Slack notifications. In October 2025 Graphite folded its Diamond reviewer and its chat product into one thing called Graphite Agent, which gives feedback in the diff, applies fixes, and lets you argue with the suggestion and iterate to a commit without leaving the page.

As of September 2026 its pricing page lists Hobby free for personal repos, Starter at $20 per user per month billed annually, Team at $40 with unlimited AI reviews and chat plus the merge queue, and custom Enterprise pricing.

If your problem is that trunk keeps breaking and your team ships in one giant branch per feature, Graphite is a better purchase than a diagram tool. I would say that out loud to a customer.

Where stacking stops helping

Stacking is an authoring discipline. Someone has to decide where the seams are, and that someone is the person writing the code.

That works when a person writes the code. It works badly when the code arrives whole. An agent opens a pull request that touches a route, a queue, a worker, four config files, a migration and eleven tests, and it did not think in stacked increments while doing it. To turn that into a stack you have to first understand it well enough to know where the seams are, which is the exact work you were hoping the stack would save you.

There is a second case where stacking does not reach: the review still happens per pull request, and a reviewer looking at the third PR in a stack usually has no picture of the first two. Small pull requests are easier to read. A sequence of small pull requests is not automatically easier to understand than one large one, because the understanding lives in the relationship between them.

Graphite changes the size of what you review. PR Lens changes what reviewing it costs you.

What PR Lens does inside a stacked workflow

Nothing special. It draws each pull request as it opens and redraws it on every push, and it does not care whether that pull request is standalone or the fourth in a stack. The hosted App keeps a baseline of the repository and tracks it across commits, so each PR is drawn against the system rather than against its own diff in isolation.

The output scales down honestly. A one-line change to one endpoint does not get a diagram pretending to be an architecture review.

That matters more in a stacked workflow than in a normal one, because a stack is a pile of small pull requests, and a tool that produced a wall of boxes for each of them would be worse than useless. Each PR in the stack gets a diagram the size of its own change, and each one links to a canvas where you can pan, zoom, and press play to walk the change one step at a time, dimming everything else and lighting the cards and routes for that step.

For the pull request that did not get stacked, the one that arrived at 400 lines across 12 files, the routine I use is in How to review a 400-line pull request in ten minutes.

Where Graphite and PR Lens overlap

Graphite Agent reviews code, so there is one place these products touch: both put something at the top of your pull request.

The difference is what that something is. Graphite Agent produces feedback: bugs, logic errors, style issues, security problems, with fixes you can apply and a conversation you can continue. PR Lens produces a picture: what the change touched, drawn against the system around it, with green for new, amber for changed, red for gone, and removed components struck through. The parts the change did not touch stay on the page.

Neither replaces the other, and they do not collide in the thread. PR Lens keeps exactly one comment per pull request, identified by a hidden HTML marker, and rewrites that comment in place rather than adding to it. It only ever adopts its own comments.

Which one to buy first

If trunk is unstable, if merges queue up behind each other, or if your team writes large feature branches by hand, buy Graphite. The merge queue alone pays for itself on a repository where CI takes twenty minutes.

If your pull requests are already small and still take too long to pick up, the problem is comprehension rather than size, and no amount of stacking will fix it. LinearB's finding that AI pull requests wait 4.6 times longer to be picked up but then finish faster than human ones once review starts points at the same thing: the delay is in front of the review. I unpacked that in Why AI-generated pull requests wait 4.6 times longer for review.

Most teams I talk to end up with one workflow tool, one bug finder and one explainer. AI code review tools in 2026, grouped by what they actually do lays out the groups with current prices, and PR Lens vs Greptile covers the finder slot, where the choice is harder than it looks.

What PR Lens will not do for you

It will not make your pull requests smaller, it will not order your merges, and it will not find a bug. It has no queue, no stacking, no severity ranking. If you install it expecting a Graphite replacement you will be disappointed within a day.

It is MIT licensed and free for open source, and it runs as a GitHub App, a GitHub Action on your own key, a CLI, or an agent skill inside Claude Code, Cursor or OpenCode. Adding it to a repository that already uses Graphite requires no configuration on either side.