It's wild, @cursor_ai with Composer 2.5 is now the optimal solution for all? of my tasks. One model, one harness, fast, efficient, reliable. Somehow achieves the impossible highest marks on best-speed + best-IQ + best-cost.
1
21
Tip: The coding models are really good at understanding "this turn" vs. "not this turn" as well as "on my next turn".
1
16
I'm finding it more enjoyable in having faster back and forth in the "early" parts of a session so that the model can rip through the rest once it understands preferences and patterns of the task at hand
1
17
Update: I've also found @mattpocockuk 's `/grill-me` skill shockingly good for this (the prompt is so simple!)
github.com/mattpocock/skills…
1
12
heya @ericzakariasson lobbing this in your direction:
Any thoughts on how Cursor might improve "findability" of past agent chats/sessions or introduce more forms of organization systems around them? Is anyone talking about this in the community or building their own solutions?
2
583
- I use multiple agent interfaces, stack ranked: Cursor (BEST), Claude Code, Codex, Antigravity
- Cursor has the best grouping/filtering so far but has an opportunity to really pull ahead by being most model-agnostic
- CC has branch status awareness/iconography baked in
1
1
66
Exploring AI tools? Sign up for Cursor, the best agent harness in town for coding, and get 50% off your first month. Try all the frontier models and see which works best for you!
Link below:
2
1
41
If you are building a vibe-coded app and you want it to be something that, if it takes off and you need to hire proper devs to manage it, I strongly recommend "the FAST Stack". Here's a prompt I ran in Codex that did a surprisingly great job... (1/2)
Install the library skills bundled with your dependencies (like FastAPI) for your coding agent 🤖
In Python or Node.js, both versions support both ecosystems ✨
github.com/tiangolo/library-…
1
1
50
.@cursor_ai has the best agent harness, hands down. I do marvel at the raw IQ of the frontier LLMs but more importantly, I trust and rely on the situational awareness and developer-EQ of Cursor's harness.
Our agent harness makes models inside Cursor faster, smarter, and more token-efficient.
Here's how we test improvements to the harness, monitor and repair degradations, and customize it for different models.
cursor.com/blog/continually-…
31
Cursor agents UI + Opus/Sonnet models is the best of everything imo, and I keep trying many other modes to continue to test and strengthen this daily reaffirmation
like most of us, I used Cursor after VS Code. then once Claude Code came out, i started using it in Warp and i've basically stuck with it for months. (tried others along the way – OpenCode, Conductor – but had issues.)
i've been back in Cursor a lot lately and man, it's so far superior to other tools – just on the AI coding experience alone.
the more i think about it, i realize it might just comes down to context management and performance. the fact i can so easily recall a conversation with rich context is so much better than the experience in Claude Code. and Cursor is so much more performant.
i dunno, having direct access to code is still such a benefit. and despite all the UI things that have driven me nuts about Cursor over the last year+, they've definitely smoothed a lot of stuff out and it's pretty hard to beat right now.
64
Ok @googlechrome you're saying "it's a skills issue"? Ouch. But, you're probably right. 😆
21
I would really love to know if anyone, anyone has truly gotten OpenClaw or Hermes working with any model besides one from Anthropic. It's just utter fails all the way down for me, I'm starting to get VERY suspicious.
22
.@SpiritAirlines loves fine print and money and hates humans apparently. I'm just glad my children are home safe after a dangerous overnight stay in an airport and not one but TWO canceled flights. Due to weather, they say now, but not what they said last night.








