You're doing code reviews wrong - AMA w/ CEO of Graphite

Hey Product Hunt!

@catt_marroll great question! One of our core theses is that the best developer experience for shipping code is a highly integrated product that spans the entire pull request lifecycle, from creation all the way through review, testing, merge, deployment, and beyond. This belief comes from our experience using incredible internal devtools at large companies like Meta and Google, both of which have built this kind of end-to-end tooling to help them ship quickly even as their eng teams grow to tens of thousands.

Everything we build at Graphite is designed to accelerate code review for high-velocity teams, whether you're 5 engineers or 5000. With that framing, stacked PRs and AI code review are highly complementary - stacking helps engineers stay unblocked and write more changes, and Diamond (our agentic AI code review companion) gives engineers instant feedback on those changes, saving hours of back-and-forth review cycles.

What you're seeing is Graphite growing into that vision for an end-to-end pull request toolchain. Graphite users can now:

• Create stacked PRs with our CLI & VS Code extension

• Generate PR titles and detailed descriptions with AI

• Assign the best reviewers with Graphite Automations

• Get instant feedback from Diamond, our agentic AI code review companion

• Stay on top of updates via our PR inbox, Slack notifications, and MacOS menu bar app

• Review changes and accept AI-generated code suggestions and CI fixes with one click

• Automate the merge process with Graphite's stack-aware merge queue

Many companies string together a bunch of home-grown and off-the-shelf tools to approximate this workflow, but as any of our customers will tell you the whole is truly greater than the sum of the parts.

@merrill_lutsky awesome, appreciate the thorough reply and really cool to get a sense of the whole product offering. honestly i didn't realize there were this many features in the graphite "toolkit".

We use diamond regularly, and occasionally stack PR's. I've never fully bought into the stacking flow, and maybe thats more of a me problem than anything, but i have gotten a lot of value from the review tool.

it seems like this flow lives "on top of" github if for no other reason than GH being a 500lb gorilla. would you consider competing with GH as a git hosting space as well? I would imagine that gives you even more levers to tighten the product integration into the dev flow.

regards for the ama, cool to see companies grow and mature.

@steveb I'm somewhat biased here, but I genuinely think that as AI code generation gets better and better, much of the job of software engineering will shift from writing code to reviewing it. We'll kick off agents to build entire features and then give them feedback on their PRs much like we do to our human engineer teammates today, and our vision is that all of this iteration and refinement happens on Graphite. More generally, if code is primarily being written by agents, it means the local development experience as we know it will become far less central to the engineering process, and interfaces for interacting with agents and reviewing their work (e.g. preview environments) become more critical as pieces of infrastructure. Even beyond that, I think we'll see increasingly small teams moving to "big company" developer toolchains (e.g. stacked PRs, large monorepos, merge queues) to handle the volume of changes they're now capable of producing with AI. This is why we're building all these features in a nicely integrated platform!

@merrill_lutsky Love it.

@aaronoleary I agree that this will likely start to limit the rate at which AI code generation improves. We're seeing this even more broadly with LLMs as we start to reach the limits of datasets like Common Crawl, and many of the model providers are starting to supplement with synthetic data. I think a lot of the innovations that we've seen recently in the way reasoning models like Deepseek have been trained suggests that reinforcement learning is the most powerful lever for improving model performance, which means that companies with data around acceptance rates of AI-generated code are going to have a big advantage in the battle for the SOTA for code generation. Beyond the immediate revenue potential, I think this is a key reason why we see Anthropic releasing Claude Code and OpenAI announcing their forthcoming coding agents.

@karan_mandal2 summarization can be hard to get right, but one of the most-used AI features in Graphite is actually our pull request description generator. Writing good PR descriptions for your reviewers can be tedious, but we've found that AI is great at summarizing the change and formatting it into a PR template (if your team uses one), which you can then edit and publish!

@steveb we’ve had a few of those moments! Graphite started as an internal tool we built for ourselves while working on a different devtools idea (instant rollbacks for native iOS app releases), and one of the first signs that we were onto something was we had a couple demos where there would be an ex-Meta eng on the other side who would say something like “okay this iOS thing is cool and all but I heard you guys built an internal thing for stacked PRs - how can I buy that?”

Our original waitlist launch was another one of those moments - we were aiming for 500 developer signups in the first 24 hours and got to over 3k! US-West-2 went completely down for a few hours the night before that launch so we were all scrambling to try to migrate to a different region to keep the launch on schedule.

A funny one that happened last year was GitHub’s API team reached out to us because we’ve now become one of and perhaps the single heaviest user of some of their endpoints as we’ve scaled.

@steveb lately I've been listening to a lot of Trym, Script, Eli Brown, HI-LO, and Reinier Zonneveld! Also love some of the more melodic stuff like Argy, Massano, Kevin de Vries... wbu?

@merrill_lutsky Thanks! I'm all melodic/minimal house sets on Soundcloud. Recent favs include this this and, because I missed Burning Man last year, this.

@lagap I think the biggest opportunity is that AI can make the outer loop simultaneously more thorough and less tedious. Previously you had to directly trade off exhaustiveness/confidence in review and testing with both human attention and compute cost. Now AI can help us quickly review and test every code change thoroughly, and the costs are rapidly falling as well. Graphite’s vision for the new outer loop is one where code changes go from being “at rest by default” (as they are today) to “in motion by default”, with AI moving them along the process, proactively fixing errors as they come up, and tagging in human engineers as needed.

@greffin_dony very few engineers would say they “love” code review, but we believe with the right workflows and great tooling we can make it a much less tedious and more helpful part of the development process!

@umair_zubair1 very true - code review and the entire “outer loop” of development gets more complex as teams grow in size and expand globally. This is why the largest software companies like Meta and Google have built amazing internal tools to make the process more efficient at scale, and that’s what inspired Graphite!