Cursor just started their war on GitHub - CourtneyR.dev's 2nd Brain

Cursor just started their war on GitHub 🔗 **Source:** [Cursor just started their war on GitHub](https://www.youtube.com/watch?v=1_P1K9i1Oq0) > [!note]+ Document Notes > tags: [cursor, github, code-review, ai-development, software-engineering] > author: Theo - t3․gg > source: YouTube > reference: https://www.youtube.com/watch?v=1_P1K9i1Oq0 > date created: 2024-12-20 12:30 > > # 🏷️ Title: Cursor just started their war on GitHub >  ## 🧰 Action Inbox - [H1] Evaluate Graphite for code review workflow — better stacked diffs for AI-generated code #code-review - [H2] Consider switching from GitHub to Graphite platform — faster UI, better PR management #github-alternative - [H3] Monitor Cursor-Graphite integration features — shared history between IDE and reviews #ai-development ## 🧭 Table of Contents - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ Paraphrase](#️-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-assessment) - [✅ Logical Validity](#-logical-validity) - [🛡️ DARVO Analysis](#-darvo-analysis) - [🎯 Conclusion](#-conclusion) ### 🔗 References - [Cursor Blog Post](https://cursor.sh/blog) - [Graphite Blog Post](https://graphite.dev/blog) - [Depot](https://depot.dev) ## 📝 Summary - Cursor acquired Graphite, a code review platform used by hundreds of thousands of engineers - The acquisition aligns with the shift from writing code being the bottleneck to reviewing code being the bottleneck in the AI era - Graphite's stacked diff approach and superior code review UI will integrate with Cursor's AI capabilities - This represents a strategic move to challenge GitHub's dominance with an AI-first development platform - All 70+ Graphite employees will join Cursor, maintaining the existing product while building deeper integrations ## 📖 Main Content Cursor has acquired Graphite, a GitHub-wrapper code review platform that has been solving workflow problems for fast-moving engineering teams. The acquisition represents a strategic alignment between AI code generation (Cursor) and modern code review processes (Graphite). Graphite emerged from ex-Meta engineers who wanted to bring Facebook's internal development workflows to the broader developer community. Their key innovation is "stacked diffs" - allowing developers to create pull requests on top of other pull requests, enabling faster iteration without blocking on reviews. The timing is significant because AI tools like Cursor have shifted the bottleneck from code writing to code reviewing. Teams that previously filed 2-3 PRs per day are now filing 10+, creating massive code review backlogs that traditional GitHub workflows can't handle effectively. Graphite's platform handles large pull requests (4,000+ lines) that would crash GitHub's interface, provides useful homepage dashboards showing pending reviews, and offers superior performance through extensive caching of GitHub's notoriously slow APIs. The acquisition enables several strategic possibilities: shared history between Cursor IDE and Graphite reviews, better AI review agents, potential migration away from GitHub's infrastructure entirely, and an integrated "create-review-merge" workflow optimized for AI development. ## 🗒️ Notes #### ⏱️ [00:00](https://youtu.be/1_P1K9i1Oq0?t=0) — [H1] Cursor acquires Graphite in GitHub challenge - Key idea: Cursor acquired code review platform Graphite to compete with GitHub - Evidence: "Looks like cursor is trying to compete with GitHub" #### ⏱️ [02:15](https://youtu.be/1_P1K9i1Oq0?t=135) — [H2] AI shifts bottleneck from writing to reviewing code - Key idea: AI tools make code generation fast but code review becomes the constraint - Evidence: "filing less than 10 [PRs per day], you're probably having a slow day" #### ⏱️ [04:30](https://youtu.be/1_P1K9i1Oq0?t=270) — [H2] Graphite enables stacked diffs workflow - Key idea: Stacked diffs allow PRs on top of PRs for faster iteration - Evidence: "you can file PRs on top of PRs and improve them as individual chunks" #### ⏱️ [07:45](https://youtu.be/1_P1K9i1Oq0?t=465) — [H2] GitHub's UI fails with large PRs - Key idea: Graphite handles massive pull requests that break GitHub's interface - Evidence: "4,000 lines of code added... this is the type of PR that would traditionally just entirely break things on GitHub" #### ⏱️ [09:20](https://youtu.be/1_P1K9i1Oq0?t=560) — [H3] Meta workflow patterns benefit broader teams - Key idea: Ex-Meta engineers brought internal Facebook workflows to external teams - Evidence: "Most of the original graphite team and founders are ex Meta engineers" #### ⏱️ [11:30](https://youtu.be/1_P1K9i1Oq0?t=690) — [H2] AI democratizes need for advanced workflows - Key idea: What used to benefit 5% of engineers now benefits nearly 100% due to AI - Evidence: "what used to be like maybe 5% of engineers would have benefited greatly from these flows, it's now much closer to 100%" #### ⏱️ [14:15](https://youtu.be/1_P1K9i1Oq0?t=855) — [H1] Cursor's editor quality issues drive acquisition strategy - Key idea: Cursor's buggy editor needs better code review to improve quality - Evidence: "Cursor got so buggy that they bought a code review company to fix it" #### ⏱️ [16:45](https://youtu.be/1_P1K9i1Oq0?t=1005) — [H2] Pull requests aren't right primitive for AI - Key idea: Traditional PRs don't capture iterative AI development workflows - Evidence: "Pull requests are just not the right primitive for AI" #### ⏱️ [19:30](https://youtu.be/1_P1K9i1Oq0?t=1170) — [H3] Integration started with API discussions - Key idea: Acquisition talks began as technical integration conversations - Evidence: "the conversation about an acquisition started way back when they were talking about potentially integrating the cursor background agent" #### ⏱️ [22:00](https://youtu.be/1_P1K9i1Oq0?t=1320) — [H1] GitHub's leadership lacks developer experience - Key idea: GitHub's leaders don't write code, limiting their developer focus - Evidence: "the person in charge of all of GitHub... he's never written code. He has no interest in writing code" #### ⏱️ [24:30](https://youtu.be/1_P1K9i1Oq0?t=1470) — [H2] GitHub API performance problems drive alternatives - Key idea: GitHub's slow, unreliable APIs force Graphite to build caching layers - Evidence: "90 plus% of the work they do on back end is just trying to make the GitHub API usable" #### ⏱️ [26:15](https://youtu.be/1_P1K9i1Oq0?t=1575) — [H1] Graphite positioned as GitHub replacement - Key idea: Graphite could become full GitHub alternative, not just overlay - Evidence: "Graphite is actually well positioned to be the real alternative to GitHub" #### ⏱️ [28:45](https://youtu.be/1_P1K9i1Oq0?t=1725) — [H2] All 70+ Graphite employees joining Cursor - Key idea: Large team acquisition maintains existing product focus - Evidence: "over 70 people at Graphite during this acquisition and all of them will be staying" ## 📚 References - [Cursor](https://cursor.sh/) - [Graphite](https://graphite.dev/) - [Depot](https://depot.dev/) - [GitHub](https://github.com/) ## 🗣️ Paraphrase - Cursor acquired Graphite to solve the code review bottleneck that AI development creates, as teams now generate far more code than they can effectively review - Graphite's stacked diff workflow and superior GitHub interface will integrate with Cursor's AI capabilities to create an end-to-end development platform - This acquisition positions the combined company to challenge GitHub's dominance by offering AI-native development workflows that GitHub's legacy architecture and leadership cannot match ## 📰 Newsletter **Big News: Cursor Acquires Code Review Platform Graphite** The AI code editor space just got more interesting! Cursor announced they're acquiring Graphite, the code review platform that's been making GitHub actually usable for fast-moving teams. **Why This Matters for WordPress Developers:** - **Faster Reviews**: If you're using AI tools to generate more code, Graphite's "stacked diffs" let you break large changes into reviewable chunks - **Better UI**: Handle those massive plugin updates or theme overhauls that make GitHub's interface crash - **Team Workflows**: Perfect for agencies managing multiple client projects with frequent deployments The acquisition signals a shift from "how fast can we write code?" to "how fast can we review and ship it safely?" - exactly the problem many WordPress teams face with AI-assisted development. **Try It Out**: Graphite works as a GitHub overlay, so you can test it on your existing repos without changing your workflow. Resources: [Learn WordPress](https://learn.wordpress.org/), [Developer Docs](https://developer.wordpress.org/), [WP-CLI](https://wp-cli.org/) ## ✅ Action Items #### DevRel - Developer experience: Create migration guides from GitHub to Graphite workflows - Education: Workshop on stacked diff methodology for AI-assisted development - Community management: Monitor developer adoption and pain points with new workflows #### OSPO - Software Inventory: Evaluate impact on code review and approval processes - Developer Support: Training on new review tools and AI integration patterns - Strategic Alignment: Assess implications for GitHub-dependent workflows #### Community - Contributors: Evaluate stacked diff benefits for WordPress core contribution workflow - Maintainers: Consider adoption for high-velocity projects with multiple contributors #### WordPress - Core: Assess stacked diff workflow for large feature development - Core Performance: Better review process for performance-critical changes - Community: Evaluate for WordCamp technical content review processes ## 🧪 Hypotheses - Teams using AI code generation tools will achieve 3-5x faster review cycles with stacked diff workflows compared to traditional GitHub PRs - Graphite's GitHub API caching will provide 10x performance improvement for large repository operations - Integration of Cursor IDE history with Graphite reviews will reduce context switching and improve review quality - Organizations will migrate from GitHub's interface to Graphite while maintaining GitHub as storage layer ### 📊 Methodology - A/B testing: Compare review cycle times between GitHub native and Graphite workflows - Performance benchmarking: Measure UI responsiveness on large PRs (1000+ lines) across platforms - User experience studies: Track developer productivity metrics before/after stacked diff adoption ### ❓ Questions - How will GitHub respond to the competitive threat from an AI-native development platform? - Can Graphite maintain product focus while integrating with Cursor's broader AI strategy? - Will other AI IDE companies pursue similar code review platform acquisitions? ### 🔮 Future Research - Impact of AI-generated code on traditional software review processes and quality metrics - Developer workflow optimization patterns emerging from AI-assisted development - Strategic positioning of development toolchain companies in the AI-first software era ### 💡 Implications - Practice: Teams using AI tools need to rethink code review workflows and tooling - Policy: Organizations may need updated code review standards for AI-generated contributions - Security: Faster code generation requires more sophisticated automated review and validation - Education: Developer training programs should include modern review workflows beyond basic Git/GitHub ### 🧠 Learning Styles Assessment Reading level estimate: 12th grade #### 👀 Visual learners - Present: Screenshots of Graphite interface, stacked diff visualizations - Improve: Architecture diagrams showing Cursor-Graphite integration flow #### 🔊 Aural learners - Present: Video format with detailed verbal explanations - Improve: Podcast-style discussion of strategic implications #### 🗣️ Verbal learners - Present: Clear analogies, technical term definitions, industry context - Improve: More comparisons between traditional and AI-native workflows #### 🧩 Kinesthetic learners - Present: Step-by-step workflow demonstrations - Improve: Hands-on tutorial for setting up Graphite with existing GitHub repos #### 🧠 Logical learners - Present: Clear cause-effect relationships, strategic reasoning - Improve: Decision matrix comparing code review platforms #### 🧑‍🤝‍🧑 Social learners - Present: Community discussion, team workflow implications - Improve: Case studies from teams who adopted similar tools #### 🧘‍♀️ Solitary learners - Present: Individual productivity benefits clearly outlined - Improve: Self-assessment checklist for workflow optimization needs #### 🌿 Naturalistic learners - Present: Evolution of development workflows over time - Improve: Ecosystem thinking about tool interdependencies ### 📊 Accessibility Assessment #### Screen reader semantics - Present: Clear headings, video has captions - Improve: More descriptive alt text for UI screenshots #### Keyboard navigation - Present: Standard video controls accessible - Improve: Timestamped navigation for key topics #### Contrast and readability - Present: High contrast video interface - Improve: Transcript formatting for better readability #### Media and captions - Present: Auto-generated captions available - Improve: Human-reviewed captions for technical terminology accuracy ### 🌍 DEIB Assessment #### Cultural breadth - Present: Focus on global developer tools and workflows - Improve: Examples from diverse company sizes and regions #### Inclusive language - Present: Generally inclusive technical discussion - Improve: Avoid casual profanity, more professional tone #### Representation - Present: Mentions diverse company users (Shopify, Figma, etc.) - Improve: Include perspectives from underrepresented groups in tech #### Psychological safety - Present: Balanced critique with constructive suggestions - Improve: Less inflammatory language about competing tools ### ✅ Logical Validity #### Strong reasoning - Deductive strength: Clear logical connection between AI code generation and review bottlenecks - Inductive strength: Good pattern recognition from multiple company examples - Causal reasoning: Well-supported claims about GitHub's performance limitations - Analogical strength: Effective comparison to historical workflow evolution #### Fallacies - Appeal to authority: Over-reliance on personal relationships and insider knowledge - Ad hominem: Attacks on GitHub leadership rather than focusing on product limitations #### Tone of voice - Overall tone: Enthusiastic but biased due to financial interests and personal relationships ### 🛡️ DARVO Analysis - Deny: No significant denial patterns observed - Attack: Some attacks on GitHub's leadership and technical capabilities - Reverse victim/offender: No reversal patterns detected - Emotional cues and rhetoric: High enthusiasm potentially influenced by financial stake in outcome ## 🎯 Conclusion The Cursor-Graphite acquisition represents a significant shift toward AI-native development workflows that could fundamentally challenge GitHub's dominance. By combining advanced code generation with sophisticated review processes, this merger addresses the core bottleneck facing development teams in the AI era. The strategic alignment makes sense both technically and business-wise, though success will depend on execution and maintaining the quality that made both tools valuable independently. For WordPress developers and other fast-moving teams, this could provide the workflow improvements needed to effectively leverage AI assistance while maintaining code quality and team collaboration. ## 🧱 Refactor Appendix ### 📚 Literature Split Pad ### [H1] AI shifts development bottleneck to code review > [!quote] [Cursor just started their war on GitHub](https://www.youtube.com/watch?v=1_P1K9i1Oq0) by [[Theo - t --- ## Full Transcript [00:00] Only 10 days left till the end of the [00:01] year. I'm sure the news cycle is going [00:03] to finally slow down and we can all take [00:04] a break. Right about that. Oh boy. Looks [00:10] like cursor is trying to compete with [00:11] GitHub. I don't even really know where [00:13] to start with this one because I I have [00:16] too much context. I know this is kind of [00:18] a meme at this point, but I've been [00:20] close with Graphite for years. I'm an [00:21] early investor in Cursor. I spent hours [00:23] at the cursor office just a few days ago [00:26] talking about a lot of problems that I'm [00:27] sure are going to come up throughout [00:29] this video. I have a lot of feelings [00:31] about this. The TLDDR is that GitHub is [00:34] [ __ ] The long version is going to [00:36] come right after the break, but to just [00:38] give you an idea, Graphite has made code [00:41] review and managing all of your poll [00:43] requests significantly better. And [00:45] aligning with Cursor is going to set [00:47] them up for a really, really bright [00:49] future. Graphite has sponsored videos in [00:51] the past, as you can probably guess. [00:52] They are not sponsoring this one. I'm [00:53] taking no money from any of these [00:55] companies to cover this. They don't even [00:56] know I'm doing it. Okay, Graphic kind of [00:58] knows. I did a call with the CEO earlier [01:00] just to get his perspective on things. [01:01] And I can't wait to share all of the [01:02] things we talked about right after a [01:04] break from today's actual sponsor. Every [01:05] once in a while, I find myself [01:07] reminiscing on the old days where filing [01:09] two to three PRs in a given workday was [01:11] considered really, really productive. [01:13] Now, if you're filing less than 10, [01:14] you're probably having a slow day, [01:16] especially with all these AI tools [01:17] making it easy to push more and more [01:19] code. But there's a downside to this. [01:21] And no, it's not just the quality of the [01:22] things that we ship. It's the amount of [01:24] time we're spending waiting. I've never [01:26] spent such a huge portion of my life [01:28] just sitting there waiting for my CI to [01:30] build, for my Docker images to compile, [01:31] and all these other things that are [01:33] inherently happening whenever you make a [01:35] new pull request. Especially if you go [01:36] and make changes after, you have to wait [01:38] for the whole thing to run again. It [01:40] just sucks. Thankfully, Depot's here to [01:42] make it way, way better. They'll make [01:44] your Docker image builds way faster and [01:45] your GitHub actions, and setting it up [01:47] couldn't be easier. Before we go [01:49] further, I just need to make sure you [01:50] understand how much faster things are. [01:52] Post HTO's Docker builds went from 157 [01:55] minutes down to five. That's a 30x [01:58] improvement. The Masttoon builds went [02:00] from 16 minutes and 30 seconds to 2 [02:02] minutes and 48. That's a 6x win. They've [02:05] already been adopted by tons of huge [02:06] teams from Post Hog to Jane to AppSmith [02:08] and more. And man, the Post Hog guys [02:10] can't stop talking about how great their [02:11] experience has been. Their build times [02:13] are down by 55x and their shipping now [02:16] feels weirdly fast with Depot. Also of [02:19] note is that Ben, my channel manager and [02:21] also engineer and YouTuber, you've [02:22] probably seen this stuff before. He has [02:23] been working on rebuilding all of our [02:25] internal tooling for managing sponsors [02:26] and depot is who we chose to do all of [02:28] our builds and Docker images for it. And [02:30] you can see how complex depot makes [02:32] things. You specify what it runs on [02:34] which is the depot ubuntu image. You [02:36] then use depot build instead of docker [02:38] build. And that's pretty much it. Yeah. [02:41] So complex, right? In the time this ad [02:43] took, you could have made the switch. Go [02:45] give them a shot now at soyv.link/depo. [02:47] Before we dive in further, I should [02:49] probably explain what graphite is for [02:50] those who haven't seen the previous ads [02:52] or my previous coverage. Graphite is how [02:54] we do our poll requests for most of the [02:57] products that we build. You might be [02:58] confused cuz you know that we use GitHub [03:00] and we do use GitHub. We use GitHub a [03:03] lot, but we also use graphite as a layer [03:05] on top of GitHub to make it functional [03:08] for teams of our size. We ship very fast [03:10] at T3 Chat, like many PRs a day for many [03:13] people. Any engineer on the team could [03:15] very quickly ship like a dozen PRs. It [03:18] happens all the time. And here's one for [03:19] our wrapped feature that is hopefully [03:21] coming up by the time this video is [03:22] live. With 4,000 lines of code added and [03:25] 132 removed, this is the type of PR that [03:27] would traditionally just entirely break [03:30] things on GitHub. And here we see this [03:32] loading bar because they're going out of [03:34] their way to make the UI actually usable [03:36] for large poll requests like this. They [03:38] also have this idea of stacking and [03:40] stacked diffs which is an essential [03:43] change in how building works that they [03:45] got from their history at meta. Yeah, I [03:48] probably should have mentioned that [03:49] before. Most of the original graphite [03:50] team and founders are ex Meta engineers [03:52] that missed the way that things worked [03:54] at Facebook when they were doing changes [03:55] and wanted to try and bring the benefits [03:57] to the rest of the world by building it [03:58] on top of GitHub. All of this is in its [04:00] current state a GitHub wrapper to put it [04:03] simply. They're using the GitHub APIs, [04:05] your GitHub repos, your git histories, [04:06] and all of these things in order to make [04:08] a much better experience on top. And a [04:11] lot of legit huge companies have made [04:13] the move over to graphite because it is [04:15] so much better. And man, it really is so [04:17] much better. [04:21] I scroll the code. Like here, I'm in a [04:23] Firefox based browser, which definitely [04:25] makes things a little worse, but it can [04:27] actually handle this gigantic pull [04:29] request that adds a more than a dozen [04:31] files and touches many, many more. This [04:33] is the type of thing that would [04:34] traditionally just make GitHub collapse [04:36] and they handle it fine. The stack diffs [04:39] thing in particular is really really [04:40] really nice as well because it means [04:42] that you can file PRs on top of PRs and [04:44] improve them as individual chunks which [04:46] is great when your teams are moving [04:48] really fast and I'm now three changes [04:50] past where you're reviewing. It allows [04:52] for you to move quickly without blocking [04:54] based on reviews in a way that is just [04:56] better when you move fast. And this made [04:59] a lot of sense for these particular [05:01] archetypes of teams where this was a [05:02] problem. Specifically, medium plus-siz [05:05] teams with fastmoving engineers [05:07] benefited greatly from Graphit's [05:09] patterns. They made it way easier to [05:11] solidify review for a chunk of a change [05:13] because maybe I make a pull request or a [05:16] stacked diff in this case that [05:18] introduces a new pattern for managing [05:20] features on the back end and then when [05:22] my team's reviewing I add a new feature [05:24] and then when the team's reviewing that [05:25] I add a third one. Now, if they approve [05:27] that first PR, they have some changes [05:29] for the second, and they approve the [05:30] third, and they're all stacked on top of [05:32] each other, I can just go in that middle [05:33] slice, make changes, and once all three [05:35] of those things in the stack are [05:37] approved, I can squash the whole thing [05:38] in at once. It's a way better workflow [05:40] when you're moving fast. Historically, [05:42] that meant you needed engineers moving [05:44] fast enough to need this. It's something [05:47] I've needed for a while because my thing [05:48] was shipping really fast. I'm pretty [05:50] good at writing good code. I'm pretty [05:52] good at code review and these other [05:53] things, but my strength was always the [05:54] speed at which I could make a good [05:56] solution. Something I was really proud [05:58] of that is no longer that novel a thing. [06:00] Lots of people can just spin up Opus 4.5 [06:02] and move roughly as fast as I used to, [06:04] which is really really cool to see. In [06:06] many ways, they're actually faster, [06:07] which is still just mind-blowing to me. [06:10] But as such, what used to be like maybe [06:13] 5% of engineers would have benefited [06:15] greatly from these flows, it's now much [06:17] closer to 100%. At the very least, all [06:19] the people who are using tools like [06:21] Opus, like Cursor, that are generating [06:24] way more code, adding way more features, [06:26] benefit greatly from better code review [06:28] processes that will allow them to [06:30] improve things in chunks in ways that [06:32] moves your team much faster. And I've [06:34] seen the difference once we adopted [06:36] graphite at T3 chat and at the like [06:38] whole ping business. It is significant [06:40] and it is awesome. And when your team [06:42] does actually move that fast, tools like [06:44] graphite will make it much easier to [06:46] stop blocking in the review phase. The [06:48] harsh reality is that it doesn't matter [06:49] how fast you're shipping code if you're [06:51] not able to review it fast enough. And [06:53] this is where the cursor part comes in. [06:55] I think obviously cursor has been [06:57] essential to this idea of like shipping [06:59] way faster and building way more [07:00] features. You can even see it in their [07:02] editor which feels like it's slowly [07:04] falling to pieces. This is why I went to [07:06] their office with my team on Monday to [07:08] just sit with them and complain for like [07:10] 2 hours straight about all of the bugs [07:12] that we've been experiencing. and [07:14] they've now committed to not focusing on [07:16] new features for a bit and just going [07:18] hardcore in bug fixing mode. Every day [07:20] they've been giving my team status [07:21] updates on all the bugs they fixed, all [07:23] the new ones they notice, and what [07:24] they're doing about it. They're working [07:26] really hard to take care of these [07:27] things. And while this is meant to be a [07:29] joke, like, let's be clear, there is [07:31] some truth to the silly thing I'm about [07:33] to show. I made this dumb joke earlier, [07:36] but the best jokes have a little truth [07:38] to them. Cursor got so buggy that they [07:41] bought a code review company to fix it. [07:43] Like, yeah, it's a funny joke because [07:45] better code review tools will make it a [07:47] lot easier for a company like Cursor [07:48] shipping at that velocity to actually [07:50] make sure the code is good before it [07:52] does the next step. But also, graphite [07:55] is absurdly polished. It is such a big [07:59] surface to have as few bugs as it does. [08:02] Like, think about how many pieces exist [08:05] in a view like this. getting all of the [08:08] people who are reviewing this code, [08:09] getting the description of this feature, [08:11] getting the stacks and the related PRs, [08:13] getting all of the code in all of the [08:15] languages and all of the syntax [08:16] highlighting working in a view that [08:18] could have hundreds of thousands of [08:20] lines of code in it. These are [08:21] non-trivial problems and unlike cursor, [08:23] they're not just forking VS Code and [08:25] copying the existing implementations of [08:26] a lot of them. They have to build so [08:28] much of this themselves from scratch. [08:30] And not only have they done it, they've [08:31] done it exceptionally. My experience [08:33] using graphite has always been great. [08:35] Honestly, even though the stack diff [08:37] stuff and their CLI and everything's [08:38] incredible, just their code review [08:40] platform, like the website is so far [08:43] ahead of where GitHub is that I find [08:46] myself going to it relatively often or [08:48] just changing the GitHub review URL to a [08:50] graphite one, which will bring you [08:51] there. It's it's great, but obviously [08:54] that's not why they're getting bought by [08:55] Cursor. Nobody's buying a company for a [08:58] better view on an existing product. [09:00] There have been a couple things cooking [09:01] at Graphite that make this alignment [09:03] make so much more sense. The stack diffs [09:05] thing's absolutely part of it. Pull [09:07] requests are just not the right [09:09] primitive for AI. Things are being built [09:11] in a very different much more iterative [09:13] way than a pull request can possibly [09:15] encapsulate. So stack diffs as a new [09:17] model for shaping those changes makes a [09:20] ton of sense. But there's other things [09:21] they've been doing. Like you might see [09:23] one of them right here in the UI, the [09:26] graphite agent. Obviously, there's a lot [09:30] of effort going into trying to automate [09:31] the process of reviewing code. And I've [09:34] even said myself, as great as AI is at [09:36] writing code, it's arguably even better [09:38] at reviewing it because if AI makes a [09:39] mistake during the review, it's not that [09:41] big a deal. And if AI catches something [09:43] you might not have, it is a really, [09:45] really useful thing. When AI is writing [09:46] code, it kind of changes that dynamic in [09:48] a way that isn't as obviously inherently [09:50] positive. But letting an AI agent do a [09:52] code review before you make your team do [09:54] it can be so so nice. which is why we [09:57] have code rabbit and grapile working on [09:59] this project and also the graphite agent [10:01] and bugbot by cursor as options too. [10:04] That said, either of those options feel [10:07] particularly competitive right now. I've [10:09] been saying this to the graphite team [10:10] forever and trying to not talk about it [10:13] too much with the cursor team, but I [10:15] have just not found the quality of code [10:16] review in either of these tools to be [10:18] particularly great. And Graphite [10:20] themselves have said they're not an AI [10:22] company. They're a company making it [10:24] easier for AI developers to review their [10:25] code. They're really deeply meanted to [10:28] be a code review platform, a different [10:30] system and a different interface, a [10:32] different everything for doing code [10:34] review better and more effectively. And [10:36] they've killed it in that camp. But [10:38] their attempts to build agents into it [10:40] less their strong suit. And there's a [10:43] very very bright future here where an [10:45] alignment between the graphite team and [10:46] the cursor team would allow for code [10:48] review agents to be much much more [10:51] tightly knit. It's my understanding that [10:53] the whole conversation about an [10:54] acquisition started way back when they [10:56] were talking about potentially [10:57] integrating the cursor background agent [11:00] into graphite. One of the things [11:01] graphite was really striving to do was [11:03] be a platform where you could use many [11:05] different agents like you can spin up [11:06] clawed code to close an issue for you or [11:09] make a quick change to a PR or have [11:12] codeex go do something else or the [11:14] cursor background agent tools as well. [11:16] They were talking with cursor about [11:17] using their APIs to integrate this and [11:19] the conversation got deeper and deeper [11:20] until they eventually realized how deep [11:22] this alignment was. And also they were [11:24] talking about what it would look like to [11:25] share history between cursor and [11:27] graphite. Like when I file a PR on [11:30] graphite from changes I made in cursor, [11:32] maybe it could maintain the chat history [11:34] so you could see how and why I made [11:35] those changes. Maybe the stack diff [11:37] could align more closely with the [11:39] history within cursor, which would make [11:41] a ton of sense. It'd be really, really [11:43] nice. These are all the things I was [11:44] talking about with Merrill, the CEO of [11:46] Graphite when we did a call earlier [11:47] about it, and he just seemed genuinely [11:49] excited. He also wanted to insist that [11:52] this acquisition does not mean Graphite [11:54] won't support other tools in their web [11:56] platform. He very much still intends to [11:58] support stuff like Cloud Code within [12:00] Graphite. This is more meant to [12:02] strategically align the business with a [12:05] really strong AI partner that can keep [12:07] them moving forward without having to [12:08] worry about the next fund raise or being [12:10] profitable or all these other things [12:12] that can distract you from the mission [12:13] but also are necessary to run a [12:15] successful business. By letting Curser [12:17] and specifically Michael and crew run [12:19] the business side as he assembles this [12:22] crazy team of some of the most talented [12:24] people in the world to build these [12:25] different pieces. It makes a ton of [12:27] sense. And since the current bottleneck [12:29] they're experiencing from what I've seen [12:31] is maintaining quality when they're [12:33] doing this level of shipping. Having a [12:35] company that's whole thing is [12:37] maintaining quality and reviewing it [12:39] makes so much sense. The strategic [12:40] alignment here is nuts. Even more so [12:42] than the Super Maven acquisition, which [12:43] tragically Super Maven's no longer [12:45] really being maintained. Apparently the [12:47] people who are subbed are getting it for [12:48] free now. That's what I've been hearing [12:49] recently. But like I was hoping Super [12:51] Maven would be maintained better. Sad to [12:53] see it isn't. But also the guy who made [12:55] most of it, Jacob, has been slaughtering [12:58] at cursor. He's the one responsible for [13:00] almost all of the crazy tabcomplete [13:02] stuff that makes cursor so cool, even if [13:03] you're not using the agents. And he [13:05] built a lot of the composer one model, [13:07] which is one of the craziest code models [13:09] I've ever used in my life. Like he's [13:11] very happy to be working on those [13:12] problems. And I don't think this will be [13:14] the same case for graphite. It's my [13:16] understanding that there was over 70 [13:18] people at Graphite during this [13:19] acquisition and all of them will be [13:20] staying as part of it. They'll be [13:22] keeping their original office in New [13:24] York, which happens to be across the [13:25] street from the Cursor New York office, [13:27] which is pretty cool. They will be [13:28] continuing to focus almost entirely on [13:30] their existing product. It's just going [13:32] to be more strategically aligned with [13:33] Cursor and their goals. And I really [13:35] hope the result of this is a huge [13:37] improvement in the quality of what we're [13:39] seeing from Cursor because they really [13:41] need to get on top of stuff. if they [13:43] want to prove how great graphite is now [13:44] that they're part of their like team. [13:46] The best way to prove it is to ship [13:48] quality consistently and to no longer be [13:50] shipping so many bugs. So, cursor, hi [13:53] guys. I know you're watching. I know [13:55] we've been talking a lot on Slack and I [13:57] appreciate y'all immensely. You now have [13:59] a much stronger economic incentive to [14:02] stop [ __ ] up. Prove graphite is great [14:05] by using it to keep your [ __ ] from [14:07] regressing so much. If you can actually [14:09] tell the story early next year about how [14:11] graphite made it less likely that you [14:13] ship bugs and helped you build a much [14:15] more strong, stable, reliable platform, [14:17] that's a really really good story to [14:19] tell that'll make a graphite product a [14:20] much more compelling thing for people to [14:22] adopt. The incentives here are great. [14:24] You just need to do it. Stop adding new [14:27] bugs to my [ __ ] editor, please. Let's [14:29] see what they had to say about all of [14:30] this. I have both of the blog posts from [14:32] Graphite and from Cursor pulled up. This [14:34] is the cursor one. The way developers [14:36] write code looks different than it did a [14:37] few years ago, but reviewing those [14:38] changes, merging them safely, and [14:40] collaborating on them has increasingly [14:41] become the bottleneck for building [14:43] production grade software. The team at [14:45] Graphite has spent the past few years [14:47] thinking deeply about these workflows, [14:48] and they've built a code review platform [14:50] used by hundreds of thousands of [14:52] engineers at top organizations. Boundary [14:55] between where you write code and where [14:56] you collaborate on it feels increasingly [14:59] arbitrary, and there's a lot we think we [15:01] can build by collapsing the distance. I [15:03] absolutely agree. That reminded me of [15:05] one other interesting thing. One of the [15:07] changes graphite's been working on [15:09] recently is letting you mirror your [15:11] repos on Graphites infra. They have been [15:14] dealing with so many problems with the [15:17] GitHub API. Like 90 plus% of the work [15:19] they do on back end is just trying to [15:21] make the GitHub API usable, caching [15:23] responses so things can load fast. Like [15:25] they'll hit an endpoint and it just [15:26] takes 30 seconds to respond and have [15:28] like a 10% error rate. It's crazy what [15:30] they deal with. And more and more [15:32] they're moving things onto the graphite [15:34] platform where they can because GitHub [15:36] is slowing them down so much. If only [15:38] GitHub was built on an open distributed [15:40] standard or something to make this easy. [15:42] Oh yeah, Git. Cool. Case in point, [15:47] Graphite is actually well well [15:49] positioned to be the real alternative to [15:51] GitHub. both because it's so deeply [15:54] integrated with GitHub that you can [15:55] seamlessly work on one project across [15:57] both and have every comment, every PR, [16:00] every issue, everything synchronized [16:02] between the two because they do actually [16:04] live on GitHub. Graphite's just an [16:05] interface to it that is really well [16:07] designed, really well cached, and really [16:08] well built. Over time, I see a future [16:11] where we just use graphite and don't [16:13] bother with the GitHub part. Maybe you [16:16] initialize a git repo on your computer [16:17] for a project you're working on with [16:19] cursor and when you hit approve it will [16:22] autosync that to graphite with a really [16:24] small simple manager for that repo and [16:27] now you have an interface where you can [16:29] send a link to the thing you just did in [16:30] your chat to a friend that can now look [16:32] at it because it's hosted on graphite. [16:35] The opportunity for all of this is [16:37] crazy. And the idea that you have to go [16:39] make a repo on GitHub and hit the seven [16:42] to 13 different flags when you use the [16:44] GitHub CLI to push or you can just hit a [16:47] button in cursor and now this thing is [16:48] available to be shared. It's I am so [16:51] excited about the potential here. The [16:52] more I think about it, the more hyped I [16:53] get. Like yeah, I know I'm an investor [16:55] in cursor and I have been shilling [16:57] graphite for three plus years now. [17:00] Obviously, I'm gonna be the person who's [17:01] hyped here. But also, this has so much [17:04] potential, guys. I am I am thankful this [17:07] happened the same week that all of my [17:08] frustration with GitHub did because this [17:11] is the happy ending in so many ways. I [17:13] expect some really cool change. It's [17:15] stupid, but like they have a homepage [17:18] view that's useful. Here are all of the [17:20] PRs that I haven't reviewed since a [17:23] change was made that I'm expected to [17:25] look at. Even one that I did where I got [17:27] review comments that I haven't responded [17:29] to. None of the comments are resolved. [17:30] That's a really like imagine if GitHub's [17:33] homepage was this useful. Imagine a [17:35] GitHub where the engineering and product [17:37] was well aligned enough to understand [17:39] that they need to make their homepage [17:40] this useful. I just I cannot fathom [17:43] GitHub ever designing something this [17:45] focused on productivity because GitHub [17:48] is not a productive company. GitHub [17:50] doesn't ship anything real anymore. They [17:51] don't know what it looks like to ship [17:53] [ __ ] anymore. Graphite does. Cursor [17:55] does. The alignment here is insane. Nick [17:58] from one of my other favorites, browser [18:00] company, had a really good post about [18:01] this here. Curs is going to build the [18:03] new GitHub from first AI principles, [18:05] calling it now. No chance GitHub [18:07] survives and stays the most popular repo [18:08] host in the next 3 years. Classic [18:10] innovators dilemma where GitHub can't [18:11] move quick enough because they have too [18:13] much baggage in existing customers. This [18:14] isn't quite why it's happening. It's [18:16] happening because the leaders at GitHub [18:18] don't even [ __ ] write code. Like the [18:20] person in charge of all of GitHub is [18:22] also currently in charge of things like [18:23] AI foundry, half of Azure, and a bunch [18:26] of other random co-pilot [ __ ] And he's [18:28] never written code. He has no interest [18:29] in writing code. How do you expect a [18:31] company like GitHub to be successful for [18:33] developers if they don't have leaders [18:35] that care about developers because they [18:37] themselves are not? It can't happen. I [18:39] know for a fact that all of the [18:40] leadership at Graphite still writes code [18:42] every day. I hang out with one of the [18:44] co-founders all the time. Also, on that [18:46] note, one more fun fact. I'll stop [18:48] sinking too many of these in. just I [18:49] know too much. Okay, this acquisition [18:51] has significantly increased the number [18:53] of people at Cursor who deeply love [18:54] Outer Wilds, the greatest video game [18:56] ever made. If you haven't played it, [18:57] don't look anything up. Just go buy it [18:59] on Steam. Trust me, it's incredible. [19:02] Play it for 2 hours and if you don't [19:03] like it, flame me on Twitter and I'll [19:05] respond. It's It's magical. There's [19:07] nothing quite like it. The less you know [19:08] going in, the better. Let me know if you [19:09] love it as much as I do. Like every [19:11] opportunity I can to plug Outer Wilds, [19:13] not Outer Worlds, Outer Wilds. I [ __ ] [19:15] up the founding team at Graphite by [19:17] making them play this when they were [19:18] visiting the city and were hanging out [19:19] of my apartment. Sorry, Tomas, I had to [19:21] bring this up. It's This game hits a [19:23] certain type of nerd really, really [19:25] hard, and the graphite team is that type [19:27] of nerd. And as silly and arbitrary as [19:29] it is, it makes me way more excited [19:30] about them being part of the editor that [19:32] I use every day. It's just a it's a good [19:35] alignment. I'm hyped about this. Let's [19:37] see what they have to say quick. I [19:38] haven't even read this post yet. [19:40] Obviously, theirs has a lot more detail [19:41] because they're the review company, not [19:43] the feature company. Makes sense. [19:45] They're all to announce that Graphite [19:46] has signed definitive agreement to join [19:48] Cursor. This deal brings together the [19:50] best-in-class tools for creating, [19:51] reviewing, and merging code in the age [19:53] of AI. It's funny that only one of these [19:55] three things is what Cursor does. The [19:57] other two are what they do at Graphite. [20:00] But yeah, more of the process is owned [20:03] by graphite than cursor, as silly as [20:04] that sounds. You might spend more time [20:06] in your editor, but more of the steps [20:07] exist outside of the editor, and [20:09] graphite has done crazy things to [20:11] improve those. Tomas, Greg and I started [20:13] building graphite almost 5 years ago [20:16] because the world's best engineering [20:17] teams demanded a better code review [20:19] platform. Yep. What we never expected [20:21] was how dramatically tools like cursor [20:22] would transform our industry, empowering [20:24] engineers to build features in hours or [20:26] minutes that would have typically taken [20:28] days or even weeks without AI. [20:30] Previously, we were limited by how [20:32] quickly we could write code. But now the [20:34] bottleneck is how quickly we can review [20:35] it. Yep. I This part's funny. Like this [20:38] was my experience before all of this AI [20:41] stuff and now lots of other teams are [20:43] dealing with the thing I've had to deal [20:44] with most of my career which is me and [20:46] my co-workers ship too fast to actually [20:48] merge and now stuff like graphite made [20:51] life much easier and now lots more [20:53] people are experiencing these problems [20:54] so we all get to enjoy graphite together [20:56] the same way we like to enjoy things [20:58] like convex or terraformer and these [20:59] other tools that were good before AI but [21:02] now that AI and fast devs have so much [21:05] overlap in how they think and work like [21:07] they prefer prefer things to be in the [21:08] codebase, not in a dashboard. They [21:10] prefer review processes that are faster [21:12] and have tighter loops and cycles. All [21:14] these things that used to only matter to [21:16] top 1% and 5% devs suddenly matter to [21:19] the whole industry. There are so many [21:21] companies already using stuff like [21:22] graphite. They list a couple of the big [21:24] ones here. Leading engineering orgs like [21:26] Shopify, Snowflake, Robin Hood, Figma, [21:28] and Ramp have realized the age of AI [21:30] demands an entirely new dev tool chain, [21:32] and these companies have turned to [21:33] graphite to help them review and merge [21:34] code at the new speed of creation. I [21:36] also know Versell uses them heavily too. [21:38] Listening to our customers journeys with [21:40] AI adoption has helped us sharpen and [21:41] refine our ultimate vision for graphite. [21:43] One integrated platform where humans and [21:45] agents can create, review and merge code [21:48] changes collaboratively. A future where [21:50] PRs become self-driving, where the inner [21:52] and outer loops collapse into one [21:54] iterative process, and where speed and [21:56] quality are no longer at odds. But [21:58] realizing that vision required deep [22:00] integration with the tools our customers [22:02] use to generate code and naturally [22:04] cursor was our first stop. As I [22:06] mentioned before they were talking about [22:07] these integrations both integrating [22:09] different background agents into [22:11] graphite's background agent tools as [22:13] well as integrating history and sharing [22:15] between cursor and the PR review [22:17] process. And that led to this crazy [22:20] acquisition happening now. For our first [22:22] conversation with the cursor team we [22:23] were deeply impressed by their [22:25] thoughtfulness, ambition and craft. We [22:27] quickly found that we had a lot more in [22:29] common than just our largest customers [22:30] and investors. We shared both a nearly [22:33] identical vision for the future of [22:34] software development and set of values [22:37] around talent density, urgency, and [22:39] craft. An acquisition wasn't on our [22:41] radar. Graphite's growth curve, product, [22:43] and team are stronger than ever, and the [22:44] future of code review has never been [22:46] more exciting. I was just hanging with [22:47] these guys at the Excel office a few [22:49] months ago, and they I think they had [22:51] just done a raise if I recall. can't say [22:53] for sure. So, I'm not leaking because I [22:55] don't actually know. But, it shows this [22:58] was definitely not what was being [23:00] planned. Funny enough, this acquisition [23:02] was leaked to me from another friend at [23:04] another competitor just a few days ago, [23:07] and I hadn't talked to either party [23:08] about it. I just assumed it was a weird [23:10] one-off rumor that there was no way it [23:12] was going to happen because otherwise it [23:13] would have made it to me. And then it [23:14] did this Friday afternoon. So, here we [23:18] are. But, yeah, I I did not think this [23:21] would happen. I'm sure they didn't [23:22] either, but that's how these [23:23] conversations go. You hit somebody up [23:25] for one thing and you get something else [23:27] entirely different. There's a joke in [23:29] the VC world, which is if you want [23:31] advice, ask for money. And if you want [23:33] money, ask for advice. If you want to be [23:35] acquired, ask for partnerships. And if [23:36] you want a partnership, ask to be [23:38] acquired. Seriously though, it's [23:40] hilarious. We chose to join forces with [23:42] Cursor because of the opportunity to [23:44] create something phenomenal together. [23:46] The endto-end platform for building with [23:48] AI. We've long dreamed of connecting the [23:49] surfaces where we create, collaborate, [23:51] and validate code changes. And this deal [23:53] dramatically accelerates the timeline on [23:55] which we can make this a reality. Now, [23:57] the important piece, what this means for [23:58] graphite. If this ends up being the [24:00] death of graphite, I will be the single [24:02] most upset, loudest complainer about [24:04] this imaginable because I re I really [24:07] need graphite. Cursor is the best AI IDE [24:10] by far. But you can do a lot with VS [24:13] Code plus C-Pilot or VS Code plus using [24:16] cloud code in the terminal or wind surf [24:18] or I can't even in good faith say [24:21] anti-gravity but it feels like the gap [24:23] between cursor alternatives and cursor [24:26] it's real but it's not that big. Whereas [24:29] the gap between graphite and its [24:30] competitors is I don't have any interest [24:33] in using any of the competitors. I've [24:35] tried a lot of them. I've certainly [24:37] spent a lot of time on GitHub and you'll [24:39] pull Graphite from my cold dead hands. I [24:42] rely on it immensely for doing my job [24:44] and running my company. I can somewhat [24:46] trivially see myself switching idees. I [24:49] cannot fathom switching off of graphite. [24:52] If they take it from me, there will be [24:54] hell to pay. Hold me to that, chat. I [24:56] need y'all to make sure we keep graphite [24:59] aligned because if we lose them, GitHub [25:02] wins. And nobody wants GitHub to win. [25:04] even GitHub, judging by how they run it [25:06] recently. Graphite's product and brand [25:08] aren't going anywhere. Fingers crossed. [25:11] We'll continue to be the place where [25:12] hundreds of thousands of engineers at [25:14] the world's top software companies [25:15] review and merge code. The only change [25:17] is that we now have far greater [25:18] resources to continue to deliver an [25:20] incredible product. In the coming [25:21] months, you can expect us to do things [25:23] like double down on our best-in-class [25:24] stack PR's platform and merge cues, ship [25:27] thoughtful integrations between cursor [25:28] and graphite to seamlessly connect local [25:30] dev, background agents, and poll [25:31] requests. Leverage Cursor's expertise in [25:34] coding models to make Graphites AI [25:35] features even more intelligent. Combine [25:37] the best of Graphites AI reviewer and [25:39] Cursor's Bugbot into the most powerful [25:41] AI reviewer on the market. Good luck [25:43] there. Uh you have some strong [25:45] competition, many of which sponsor me. [25:47] Even if they don't figure that out [25:48] though, if they figure out everything [25:49] else for the platform side, they will [25:50] still win. And then laying the [25:52] foundations for an even bolder product [25:53] vision, which we'll share more on soon. [25:56] I think they're still figuring that part [25:57] out, to be honest. But there's so much [25:59] opportunity here for everybody to be [26:01] aligned and work together. I'm excited. [26:03] Even if this is just a team that's good [26:05] at code review, shows up, helps cursor [26:07] polish their [ __ ] and give them a [26:09] process to keep them from shipping bugs [26:10] constantly. This was worth it. If it [26:12] ends up being a real change in the [26:15] process of I have this idea for a [26:17] change, I'm going to go build the [26:18] change. I'm going to put up the change [26:19] for review and then I'm going to merge [26:20] the change. If they can streamline that [26:23] whole process, [26:24] that could be huge. I bet they'll buy [26:27] linear next. Anyways, the transaction is [26:30] still subject to customary closing [26:32] conditions. We anticipate closing in the [26:33] next few weeks, and until then, it's [26:35] business as usual. On closing, the [26:37] entire graphite team will join cursor to [26:38] continue building. Yeah, it is kind of [26:41] crazy that they announced this agreement [26:43] before the deal went through. Usually, [26:45] they wait a bit longer, but this was [26:47] leaking. They wanted to get in front of [26:48] the leaks. Makes sense. They have a [26:51] really nice, sweet, heartfelt thank you [26:52] at the end here. Really nice to see. [26:55] Appreciate you a ton, Mel and Tomas, and [26:57] also Michael over at Cursor for seeing [26:59] the same thing I've seen in the graphite [27:01] team forever now. I've been going out to [27:04] lunch with these guys and hanging with [27:05] them since 2022, if I recall. They are [27:08] one of my favorite companies in the [27:10] whole tech world. [27:12] I am admittedly still a little scared. [27:14] This is a huge change that I did not [27:16] expect at all, but at the same time, I'm [27:19] really excited. There's a ton of [27:21] potential here. And if graphite needs to [27:23] be acquired to continue and the acquirer [27:26] isn't Microsoft using them to rebuild [27:28] all of GitHub, this is the next best [27:30] thing. You know what they say, if you [27:32] can't beat them, kill them. And it's [27:34] time for GitHub to die. I was going to [27:36] film a video about all the cool GitHub [27:38] alternatives recently, and graphite was [27:40] going to be one of them, but I don't [27:41] even know if I can at this point because [27:43] so much has changed just from here. I [27:47] think that's all I have to say on this [27:48] one. What a wild acquisition to end the [27:50] year off with. I really hope there's no [27:53] more breaking news. I [27:57] I'm so tired, but this is a good [27:59] acquisition and I'm curious how y'all [28:00] feel. Are you scared, excited? Is this [28:02] the first time you've heard of graphite [28:03] or are you going to go move off it now [28:04] that this is happening? Just let me know [28:06] how y'all feel. I am actually really [28:08] curious. I have so much insight [28:09] information here that it's hard to just [28:11] have a neutral take. So, let me know [28:12] what you guys are thinking. And until [28:14] next time, peace nerds.