843: Copilot Kills Cursor? Reacting to Github Universe Keynote

Episode Summary

Executive Summary: Recorded live at GitHub Universe, the hosts discuss the conference vibe, a Syntax meetup, and the wave of AI announcements—especially GitHub’s Copilot upgrades that bring Cursor-like features into VS Code. They evaluate model choice, repo indexing, multi-file editing, instructions, code review/workspaces, and GitHub Spark, then compare the impact on Cursor, Vercel v0, and the broader devtool ecosystem.

Main Topics: GitHub Universe atmosphere and Syntax meetup: The episode opens with reactions to being together in person at GitHub Universe, the scale of the event, and the successful Syntax meetup in San Francisco, where they met long-time online contacts and planned future meetups. AI saturation at the conference: The hosts note that nearly every booth and product at the conference is AI-related, and they discuss how quickly smaller AI tools can be ‘sherlocked’ by larger platforms like GitHub/Microsoft. Copilot gains alternatives to OpenAI models: A major announcement is that GitHub Copilot in VS Code can use alternative models such as Claude and Gemini, which the hosts view as a positive, pragmatic move that reduces lock-in and matches developer preferences. VS Code closing the gap with Cursor: They review features GitHub is adding that Cursor popularized: multi-file editing, repo indexing, inline references, intent detection, and Copilot Instructions. They note one missing standout: Cursor’s multi-line tab completion. GitHub Workspaces and code review automation: The hosts discuss a more ambitious workflow product, GitHub Copilot Workspaces, which can turn issues into plans, revise code, and potentially streamline pull requests and code reviews inside GitHub. GitHub Spark and text-to-app builders: They compare GitHub Spark to v0 and Bolt.new: a natural-language app generator that can scaffold micro-apps quickly. They debate aesthetics, simplicity, and whether these tools can compete with established low-code/no-code products. Waymo as the surprising side topic: The episode closes with both hosts praising Waymo as a standout ‘sick pick,’ highlighting the smooth experience, safety, and convenience of the self-driving ride service.

Key Arguments: GitHub is uniquely positioned to absorb Cursor-like features because it already owns the code hosting, reviews, and developer workflow ecosystem. Model choice matters less than interaction design; the real advantage comes from UI, repo context, instructions, and multi-step workflows layered on top of LLMs. Cursor remains ahead in some UX details, especially multi-line suggestions and aggressive inline assistance, but VS Code now has enough parity to pressure Cursor. Copilot Instructions and repo indexing are valuable because they let developers encode preferences and codebase context instead of relying on generic model behavior. GitHub Workspaces could shift AI from code completion to end-to-end task execution by converting issues into structured plans and implementation steps. Spark competes in the ‘build me an app’ space, but its success will depend on whether it can produce more polished outputs and integrate with existing app aesthetics. Enterprise distribution and existing GitHub lock-in may matter more than raw product quality; teams may adopt what is already where their code lives.

Data Points: Conference attendance: 3,000+ people - The hosts mention the apparent scale of GitHub Universe. Syntax meetup location: Bear Bottle Brewing at Salesforce Park - They describe the San Francisco meetup venue and surrounding park on top of a building. Wi-Fi gear setup: ~150 boxes of Wi-Fi access points - Observed in the conference back area while looking at setup gear. Time using Cursor full-time: About 2.5 weeks - One host says they have been using Cursor full-time for roughly two and a half weeks. GitHub pull requests extension downloads: 25 million downloads - They reference the GitHub Pull Requests extension in VS Code and its download count. Waymo self-driving ride count: 3 rides - One host says they have taken three Waymo rides and is now obsessed. TikTok video length: 13 minutes - A side example about a long TikTok that got major views despite low full-watch completion. TikTok views: 1.8 million views - Referenced in the long TikTok anecdote. TikTok completion rate: 0.2% watched the whole thing - Used to illustrate that distribution can matter more than completion. Shorts length change: 90 seconds to 3 minutes - They discuss YouTube expanding Shorts duration.

Pivotal Quotes: "the sauce comes from not having your model, but the sauce comes from how do you provide the data to the LLM in such a way that it actually understands it" — Host: Explaining why products win on workflow, context, and UX rather than model ownership alone. "my coding lately has just been hitting the tab key" — Host: Describing how Cursor-like autocomplete is changing day-to-day coding behavior. "I’m just a happy guy sitting here enjoying it all" — Host: Closing reflection on the rapid progress and competition in AI developer tools.

Implications: GitHub is moving AI from a coding add-on to the center of the developer workflow, putting pressure on Cursor, v0, and similar tools. Expect faster feature convergence, more model choice, and stronger competition on UX, context, and end-to-end app building.