Page 1 • Showing 20 tweets
What is a non-obvious, meaningful way that you've... 1. Automated a boring part of your daily knowledge work 2. Found an insight from messy or scattered data 3. Stumbled onto something joyful ... due to AI assistants, agents, OpenClaw, et al.? (slop bot replies pls spare me)
It’s easier than ever to write code… And yet the hard parts of software engineering are still *very hard* Don’t get discouraged by the hype. There’s still so much to learn and build.
Rules, commands, MCP servers, subagents, modes, hooks, skills... There's a lot of stuff! And tbh it's a little confusing. Here's what you need to know (and how we got here).
My year in code with Cursor! Well only my past 6 months since I got a team account, so I'm not making the top of any leaderboards haha. Whoever responds with the most tokens used building, I'll send you a bunch of Cursor credits
Ask me anything about coding with AI or Cursor!
📝 A computer for every agent What will the future of software engineering look like? Maybe we're all "builders". Maybe AI writes all of our code. It's hard to know for sure... but what I find interesting is to look at history. This idea of "build software in the cloud" has been appealing for some time. There's a number of problems we deal with to run software locally: The security model for accessing files/folders (e.g. sandboxing) The fragility of local environments (e.g. Docker, Nix, etc.) The power of the hardware (e.g. consumer hardware vs beefy VMs) The availability of the environment (e.g. laptops going to sleep) And yet... most software still happens on localhost! Why is it that even with all of these challenges, cloud computers haven't won? I've joked that this idea of a "cloud computer in a box" has been launched 100 times. It started back in the day with "dev containers" and "cloud IDEs" to remove the drudge of the four points above. Super fast hardware, always available, immediately accessible for your team. In reality, you only ended up seeing this at companies like Meta. Of course they wouldn't store secrets on their machine in .env files! They would have nice cloud environments they could connect to for everything. But this required a significant investment in local/internal DevEx. Many startups tried to bring this idea to developers more broadly, but I would argue it never really got product market fit. Including GitHub, where you can press "." to open a cloud VM ("Codespaces") with your repo. So... right idea, wrong timing. Fast forward a bit in history, and we started to get the first powerful AI models. All of a sudden there was this new demand for cloud computing. Except the models were still pretty bad! They struggled to use computers like humans. It wasn't until roughly Sonnet 4 that models started to become good at "tool calling". And tool calling is important, because it's a generic abstraction that you can train the models to be good at. You can just give them new tools! In the past year... yeah, the models have gotten much better. So why now, why is this idea of cloud computers finally starting to work? It's the combination of three major things: Better models Better products New capabilities The models can be used as effective agents. They can reliably call tools, and can be given new powerful tools. This includes teaching them to use computers like humans, navigating and clicking around autonomously. The products and agents using the models have evolved to harness this new intelligence. The ideas in the video below simply *did not work well* even 4-5 months ago. The frontier is moving fast. Going back to the local pains from the start, these secure cloud sandboxes do have an improved security model, better hardware, and are always available (your agent can run while you sleep). Software engineering is marked in capability jumps. A computer-using agent and a nice UI is helpful, but not sufficient to change the way we work (changing patterns is hard!). But new capabilities, like agents being able to test their work and send you demo videos proving it's correct, are going to meaningfully change how we build software. It was the difference internally at Cursor from ~8% of PRs merged with cloud agents to ~30% in a few months. That is staggering. Still, even with this new capability, the existing software engineering best practices remain. We're still looking at code. We're still thinking deeply about the system architecture. But it does provide a glimpse into what the future might look like. To me, that future is one with better abstractions. Removing the boring parts of building, and allowing developers to focus on bigger problems. But we have a long way to go, and so much more to build! http://x.com/i/article/20245922496367370…
All OpenAI models in Cursor are now up to 30% faster! We've upgraded all users to WebSockets with their Responses API.
It's still a foreign feeling to work at a company that is entirely meritocratic and where titles don't really matter. Everyone ships and the best ideas win. This hasn't always been the case in my career. This is also due in part from hiring strong generalists. You could call this an "AI native company" possibly, and certainly heavy use of AI for all parts of knowledge work does make it easier for a single IC to ship something end-to-end, but it's probably something deeper culturally. To be clear, we *do* still have titles at Cursor, but I now understand why other companies have singular titles like "member of technical staff". And yes, to answer the obvious question, there is always *some* power structure. If a CEO asks for something it's different than a new hire. But leaders at Cursor don't devise a grand scheme in isolation and then emerge from the shadows like "Behold! The roadmap!". I think this also has tradeoffs (there is no perfect org structure) but it is refreshingly different from how I've worked prior in my career. I don't have a grand point to make here I'm just yapping 😂
Here's how I'm coding with AI lately, might be helpful! 1. I write code primarily using agents, using models like Opus 4.5 and Codex Max for long-running tasks or tricky bugs, and Composer for frontend changes or fast updates (I still review the code). 2. Most of my web dev work happens inside the integrated Cursor browser. This is similar to using the Playwright or Chrome MCPs. Cursor can access network requests, console logs, and send elements on the page to the agent. It can also control the browser! Which can be fun for having it do automated testing: https://x.com/cursor_ai/status/199914795…… 3. I start projects pretty simple: no upfront rules, commands, or anything. As the project grows, I end up adding the most minimal versions needed. Some examples in this thread: https://x.com/leerob/status/200604326581…… 4. I heavily review all the code! Just because I'm using agents, I'm still thinking deeply about the architecture and code quality. I review the code in three passes: first, while the agent is generating it. Second, using the in-editor "agent review" before I push a PR (similar to a custom /code-review command). And finally, using Bugbot (AI code review) on my PRs. This combination helps me fix a lot of silly bugs before I ask other people to take a look at the code. 5. I always start new features with a plan (using Plan Mode). This helps significantly and I would highly recommend planning first regardless of what tool you use. I do like the Cursor UX for visualizing, editing, and saving the plans. You can view some of my plans here: https://github.com/leerob/pixo/tree/main…… 6. For really hard bugs, I use Debug Mode. It automatically instruments your app with logging, and then asks you to reproduce the issue. The agent then reads the logs and has much more helpful context to pinpoint the root cause. It also comes up with multiple theories on what the issue could be, and works through each one until it's fixed. Has been pretty helpful: https://x.com/cursor_ai/status/199882135…… 7. Always make sure you give coding agents verifiable outputs! They can't fix what they don't know about. For this, I would prefer using typed languages, and set up tools like linters and good tests. These are normal software engineering best practices, but they matter more than ever. There's also newer tools here like tsgo and biome/oxlint and bun which make dev really nice. Worth trying some of those if you do web dev. 8. I use Cursor from mobile! There's really two modes here: quick bug fixes or really big tasks. Quick bug fixes, I just pop open http://cursor.com/agents and fire away, knowing I'll get back a PR that will work 99% of the time and I can merge away. Easier than writing it down on a todo list, my PR queue is now that list. For big tasks, again I start with a plan and then I give the agent an ambitious goal (that is verifiable!). This allows the agent to run for much longer. It will keep going until it hits that goal, and if it gets lazy, you can just say "keep going" and go back to what you were doing. This is all in the cloud, in remote sandboxes, so I don't have to worry about my local machine. 9. Since someone will ask about the theme... yes I'm rocking light mode most of the time, using Cursor Light here Oh and if you're a car person... more on my car below soon
Learn about tool calling in under 3 minutes! Tools give AI models superpowers. They can read and write files, search code, search the web, run shell commands, and more.
How to use Cursor (with no coding experience)
Cursor just got a major upgrade! Agents can onboard to your codebase, use a cloud computer to make changes, and send you a video demo of their finished work. The latency of using the remote desktop is smooooth.
Recorded + edited a product demo for something new in Cursor... Launching this week! Excited for y'all to see it.
Fast/smart/cost-effective models for synchronous coding. Slower/frontier/more expensive models for the most difficult problems & background tasks that run for hours. I expect this divide to continue growing this year.
Right now everyone is trying to learn how to use coding agents. Which makes sense. But I expect that soon the meta will shift back to teaching engineering foundations. Have some interesting ideas here 😄
Cursor 2.3 is starting to roll out! We’ve spent the past two weeks fixing bugs and making reliability improvements based on your feedback. I’m so proud of our team for sprinting to make this our best release yet. We’ll be rolling 2.3 out over the week to ensure there are no regressions during your holiday coding. I also wanted to shout out @theo, who had a rough experience with Cursor and was kind enough to stop by our office and give the team a ton of feedback on how we can make the product better (including helpful reproductions of issues from their team). There are a number of bugs in this release, specifically around worktrees and plan mode, that were fixed thanks to him. Thank you! There’s also been feedback on layouts and the default panes in Cursor across X and Reddit in the past week. We’ve been digging in and listening, and I hope you’ll love where we’ve landed. It’s always been slightly annoying if you’re using Cursor with 3+ different windows open for different repos. I was talking to @schickling about this last month, and we wanted a way to better hop between different workspaces. I still think there’s more we can do here, but I really like the new Command+Option+Tab switcher to move between layouts and workspaces (thank you @JasonBud!). Since I use Command+Tab on macOS all the time, this feels very familiar. Especially when defining custom layouts like in the video below to keep my terminal always open. We’d love your feedback as we keep working to make Cursor more extensible and customizable. Another small but nice thing: we heard your feedback that it's frustrating to have releases rolled out before the changelog is live, so we made sure the 2.3 changelog was posted before you saw the update button. Speaking of the update button, while it's great we're getting updates out frequently... I also agree it can be annoying to see the update banner every time you open Cursor. We've made it so this banner will show less frequently Happy holidays everyone!
Nice quality-of-life improvement to Cursor. You can now create and save custom layouts, so Cursor will remember the exact position of sidebars, terminals, etc. We hope to make this even more customizable!
