Aaron Russell

@MyDeathMachine @zeddotdev Thank you - I hadn't noticed that option. Are there any differences between the two modes? I can't seem to select a "profile" in the traditional text thread, so is all the new agent tooling only available in the new interface?

@aaronrussell @zeddotdev You still can use the old text threads in zed with the new agent. It’s in the “…” menu in the upper right of the assistant.

Gutted that @zeddotdev have changed the assistant panel to a more traditional chat interface. I absolutely loved that the whole chat was just one big text file. No more. 🙁

Also, ⏎ now sends the message. My muscle memory would like to make a complaint.

@ryancarson @resend @SendGrid @awscloud @cursor_ai > If the LLM labs don't take the bait to pollute results... 😬 count me as less hopeful on that one

Interesting how often I use AI recommended products/services now. gemini-2.5-pro just recommended I try @resend @SendGrid or @awscloud SES This will completely replacing Google search and Google Ads. This screenshot was taken from a @cursor_ai Agent chat using gemini-2.5-pro MAX. I went on to sign up for Resend, grabbed my API key and got cranking. I even emailed the CEO to let him know why I signed up. If the LLM labs don't take the bait to pollute results based on pay-to-play, this will be game-changing. They could also just add a "Sponsored: " section at the end, but we all know where that will go ...

I use raycast on mac about a gazillion times a day. Excited to see how this transfers to ios.

🧠 Just released Anthropix v0.6. Now you can get Claude to ✨think✨ about and write a comprehensive report on the burning topic of the day: exactly how many R's are there in strawberry? github.com/lebrunel/anthr… #Claude37Sonnet #myelixirstatus

npm i agents-sdk blog.cloudflare.com/build-ai-agent…

@threepointone 👀

gm, let's ship something today

Published few words (quite a few actually) on 🌈Vibe coding™✨ ... limitations, challenges, and what it all means for crusty old devs like me 2point0.ai/posts/vibe-cod…

Everyone I know coding with LLMs seem to stick to next/react/shadcn etc. I'm wondering what people's experiences are with other frameworks, eg Vue or Svelte, or even non-JS stacks. Are the models capable enough to veer off the beaten path or are we not there yet?

@ofogelin @IterateCom @cursor_ai A terrible solution, but get claude computer use to enter "carry on" every time it sees the limit is hit.

@IterateCom @cursor_ai please make the tool call limit configurable i beg!!

🚀🚀🚀 we turned our AI coding agent “inside out”. It’s now a command line utility that any coding agent like cursor or windsurf can use to make apps in our preferred style

Watch bifbof generate the frontend of Garple entirely autonomously (at 8x), including Stripe Checkout and email. We gave it a working data model and RLS policies. It did everything else. The version from this video is now live on garple.com. Go!

@shaoruu @cursor_ai just a little thing: in chat panel, just pressing enter adds a new line, cmd+enter sends the message (or an option to toggle that behaviour)

what's one thing you really want added to @cursor_ai composer, or just to cursor in general? open to all kinds of ideas :)

@jonas Claude has a very convincing sounding answer to that question (and doesn't involve 45 minutes of waiting)

(On the phone with HMRC after holding for 45 minutes) Me: What is the difference between the Employer PAYE Reference and the PAYE Accounts Office Reference? Operator: One is the Employer Reference and the other is the Accounts Office Reference Me: Okay, but what are they each used for? Operator: Administrative purposes Me: Okay, but what specifically. For example, when would I need the employer reference? Operator: For enquiries that require an employer reference Me: Haha 😭 Operator: Haha. Is there anything else I can help you with?

Every LLM chat interface should treat "enter" as a newline. cmd+enter should send the message. Why has everyone built it the wrong way round?

@mreflow Stonks go up and stonks go down. Sometimes the market is just waiting for an excuse for a big move.

Just a little add-on based on some of the replies... Most people are saying the dip is because models can be trained with way less compute now and that's not good for NVIDIA... I believe that IS the reason for the dip. However... A few counterpoints. 1. Many analysts claim that DeepSeek either trained on much more powerful GPUs (like H100s) but can't talk about it due to restrictions OR that they started with a different set of model weights like Llama, where the expensive part of the training had already been done. This is speculation but it's pretty widespread. 2. If, in fact, it did just get much cheaper to train o1-level models with far less compute, many companies will likely still throw more compute at it. If we can train this level of model with this low of compute, image what we can train if we 10x it or 100x it. 3. If it actually is a whole ton cheaper to train new foundation models, that really means many of the big companies like OpenAI have even less of a moat than we thought. It opens the doors for many new companies and many new open-source models to pop-up, which all need compute. The lack of compute needed to train a single model seems like it will counterbalance with more companies buying GPUs because now they too can create their own foundation models specifically tailored to their needs. Just thinking out loud... My logic could be totally off. But it being cheaper to train models seems like a net win for NVIDIA. Maybe companies will buy less GPUs per company but this could be counterbalanced by a lot more companies getting into the game, due to the lower barrier of entry...

Wow... NVIDIA dropped by 17% because of DeepSeek? I wonder if the investors realize that NVIDIA and DeepSeek aren't competition. DeepSeek was trained using NVIDIA GPUs... People installing and running it locally are mostly using NVIDIA GPUs too... Scratching my head at this...

@ryancarson @cursor_ai To clarify are you saying you use o1 for the initial plan, and then other (cheaper) models to implement the plan and do the coding?

It takes awhile to get used to how to program with @cursor_ai Now I can knock out huge features in a couple hours max. This is on a big codebase too - no hello world stuff here. Here are a bunch of things I’ve learned 👇🏻

@OfficialLoganK I’d love you to try out agentflow - a js/ts framework and cli built on top of @aisdk. Its unique take is a markdown/mdx like syntax for composing workflows. agentflow.2point0.ai

What are the best agent tools / frameworks I should play around with this weekend? 🧵👇🏻