pull down to refresh

Cursor 2.0: New Coding Model and Agent Interfacecursor.com/changelog/2-0
538 sats \ 13 comments \ @k00b 18h AI
Eight agents at a time sounds like a small nightmare, but I know this is how folks are rolling now. It's like the new CPU core count.
Run up to eight agents in parallel on a single prompt. This uses git worktrees or remote machines to prevent file conflicts. Each agent operates in its own isolated copy of your codebase.
They made their own frontier model.
Introducing our first agentic coding model. Composer is a frontier model that is 4x faster than similarly intelligent models.
The in-IDE browser sounds pretty nifty.
Browser can now be embedded in-editor, including powerful new tools to select elements and forward DOM information to the agent.
There's more obviously but those are the highlights for me. I'm sensing lots of sprawl in their product and I'm worried it might not be great for writing code yourself eventually.
write
preview
related posts
209 sats \ 3 replies \ @optimism 17h
8 concurrent agents means you go from "I didn't read the output" to "There are not enough hours in a day to read all the output".
reply
100 sats \ 1 reply \ @SimpleStacker 15h
Do we know of any successful projects in production that are utilizing concurrent autonomous agents to produce and maintain their code? I'm highly skeptical about where the demand for this is coming from...
reply
30 sats \ 0 replies \ @optimism 13h
I only know of closed source proprietary software being developed like that. But in my experience, code review in many of the places that are now all-in, was non-existent anyway, so it may actually be an improvement for their product.
For open source high-qa stuff all I've seen is "if you can't maintain it, don't propose it" or in other words, lots and lots of drama.
reply
0 sats \ 0 replies \ @BlokchainB 14h
Haha
reply
111 sats \ 0 replies \ @justin_shocknet 17h
I tried the 1.7 version of the browser thing a few weeks ago but couldn't get it to work, was curious how it might be useful for scraping.
Definitely feels a little scope-creepy, they add more stuff than I can figure out how to take advantage of when I would rather simple things like the clipboard working on regular output text, not using echo to edit files half the time, and not opening concurrent shells every time it assumes it should run start or build (I have a standing prompt not to run start/build)
Feels like incentives might be starting to conflict, ultimately this is a token sales interface. Been using it over a year and have rarely exceeded the $20 plan in a given month, and the few occasions I did it was just having to deal with slower requests for a few days till cycle reset. So far this month I blew through the $20 a few times over without changing my useage.
reply
77 sats \ 0 replies \ @elvismercury 15h
Seems like they're terrified of the creeping darkness of frontier-lab monetization taking their entire market. I would certainly be afraid of that if I were them.
reply
21 sats \ 0 replies \ @nkmg1c_ventures 14h
I'm still newb tier, I'm copy and pasting occasionally from Claude and ChatGPT. I ask it about languages I don't know so I like to at least type out Rust rather than copying it all.
reply
21 sats \ 1 reply \ @nkmg1c_ventures 14h
Nice pic choice
reply
20 sats \ 0 replies \ @k00b OP 13h
I was waiting for someone to get it!
reply
33 sats \ 2 replies \ @nichro 17h
Concurrent multi-agents: Meh. Haven't found a use for it yet, just like background agents. Guess im just not at that level. It sounds like you can run multiple agents for the same prompt and compare/pick the best result, and/or have them work on different things simulatenously and then go through and approve all diffs together in a single place. Which sounds potentially neat in theory but also like a nightmare in practice, as you said.
New model: wonder how it fares against Claude 4.5 sonnet. I'm guessing Composer1 is considered a premium model too.
Embedded browser: pretty nice, I like being able to open a browser tab without needing Chrome installed.
reply
21 sats \ 1 reply \ @nkmg1c_ventures 14h
My guess is the agents will get more specialized to languages and frameworks, I could see how it would make sense on a multi-language code base or large monorepo
reply
10 sats \ 0 replies \ @nichro 13h
Good point actually. Backend agent + Frontend agent... and of course a background agent consistently checking npm libraries for vuln disclosures :D
reply
0 sats \ 0 replies \ @nkmg1c_ventures 7h
deleted by author