Charlie Duong

@benln I'd love one. What the criteria for early tab power user?

welp, that's a first

holy shit @useblacksmith is amazing

@n2parko @cursor_ai Small quality of life one: ctrl+p/ctrl+n to cycle through files and slash commands. I love that this is one of the few agent clis that supports a vim mode but was surprised to see this keybind missing.

we're about to release the next version of the @cursor_ai CLI... but before we do, we'll spend the next 12 hours fixing, improving, and shipping whatever's most important to you drop your CLI ideas, bugs (with repros), or rough edges below and we’ll ship as much as we can!

Just ralph wiggy that shit

Been using cursor bugbot for a while and it's been amazing at spotting real bugs. Paired with claude code in a ralph wiggum loop, it's been great for touching up draft PRs before adding reviewers.

Cursor's UI feels like it could potentially a great hub to do more concurrently. Something that has me coming back to the terminal frequently is how easy it is to just pop into a worktree or different project, kick off work (perhaps a few agents concurrently), then go to another dir to kick off more for a different task, review work or pop into the code to do a few edits or exploration myself if needed. I know I can do this with multiple instances of cursor or have the agent run in a worktree, but it starts to get unwieldy and resource intensive with 5 instances of lsp running. Background agents are great, but some tasks require a bit more hands on guidance or verification. Perhaps a way of quickly switching contexts/projects while maintaining a central view of ongoing a work so I can keep track of ongoing work and hop around easily when needed.

Cursor seeks to be the best and most powerful way to code with AI. What are the ways in which we could be better?

man I love how easy it is to build your own tools now

Curious where I’d be at if I didn’t switch to Claude Code for most of my work

Not bad

Okay, @cursor_ai 2.0 is fire! Composer 1, the improvements to agent interface, the review agent... It all works together beautifully. Loving the feedback loop!

woah, this is neat. Plan mode in claude code now prompts for feedback

Just got back from Japan, I miss it already. Here are a few pics from our visit to Sapporo

I didn't wake up today expecting to watch the world tram driver championship but here I am

@ninepixelgrid Damn. I love this

and we're live. 💃 commits: 1,825 lines of code added: +87,090 lines of code removed: -25,367 files changes: 906 posthog.com i hope you like it!

i've spent the last six months building the most insane thing in my career: a new PostHog website. and in about 12 hours, i'm going to push it live. but it's not just a redesign - it's like an operating system in the browser. the level of detail is insane. no normal company would actually let this get built - let alone be championed by the CEO. and it's so far "out there" that i'm actually a little nervous. will it tank signups? will people be confused because it doesn't look like a website? will it be performant? will the novelty wear off too fast? tomorrow is the day we find out. i'd say i'll be up all night stressing about it, but after 3 straight days of 14+ hours of work, my eyes are bleeding and i actually think i'll sleep like a baby. 😅 now if you'll excuse me, i've got a laundry list of bugs to squash.

Still getting great value out of Claude Code. I keep seeing a bunch of people talk about switching to Codex, may need to give it a shot. I've definitely found gpt-5 great for tricker problems, but Opus 4.1 strikes nice balance of quality and speed for me.

Jupyter notebooks in @cursor_ai feel pretty great. Been spending a good bit of time in them this past week and it's gonna be hard using them anywhere else now. Should have tried them in cursor a long time ago, but it's timely that 1.5 came with some nice improvements.