request id b0d0a721-d9ba-41d1-aebd-92cb09ca8c23.

anyone else having cursor be really lazy? how do you have it actually act like an agent?

agents are going great

Is there one quality llm frontend that supports both local models and cloud? Would use @lmstudio more if it also supported cloud models I think.

what's your setup for full-stack in cursor?.my agent frequently starts another copy of frontend / backend server, doesn't understand the vscode debugger exists, etc. guessing it's operator error😃.

ok, so there is an apple exec interview just not with gruber 🍿

RT @charles_irl: We've run thousands of LLM inference serving benchmarks at @modal_labs. We're releasing the results so you don't have to.….

it's crazy just how much craft apple puts into their icon set, nobody else goes this deep.

this is cooking actually 🧑‍🍳.

kinda stoked to pick HDR colors now, kinda scared everything is going to be garish

how are y'all using gemini 2.5 pro b/c it doesn't work at all in cursor for me

claude 4 is horrible at following attached visual designs, seen in cursor and figma's dev mode. playwright-mcp is helpful in cursor, though!.

what if we wrote our tests like this?.load_dataset("mylib/every-mylib").map(mylib.fn).

how are we optimizing gpu utilization during rl fine-tuning?.dedicated gpus for inference mean they are idle during rl training, unless you update weights during a rollout. switching between rl and inference means waiting for stragglers. ideas?.

🤦‍♂️had to make a "linked in" profile to attend the anthropic event in nyc

Just me or is claude 3.5 the only model that works reliably in Cursor?.

python packaging is still in the dark ages. you go on pypi and you can't inspect the code or files in the package, no idea what dependencies it has, etc.