All the developers complaining how the context switching of running multiple agents is wiping them out haven't had the years of TikTok training required. Skill issue.
English
Rafael Mendiola
16.6K posts
Rafael Mendiola
@GroundControl
Startup founder, making https://t.co/XQPxX0zC6g and building an AI-native company. MIT grad. React Native, AI and operational efficiency is my jam
Boston Katılım Nisan 2008
2.1K Takip Edilen1.9K Takipçiler
@rickyfm The real issue is the 2 factor friends we made along the way
English
Isn’t the real issue not using 2 factor auth or trusted publishing?
flavio@flaviocopes
How Axios was compromised 🤯
English
@intelligibabble @bcherny Thanks Leland! I really appreciate the clarity. I'd love to show you how it's evolved at some point
English
@GroundControl @bcherny From your description, it should be allowed. See here
Boris Cherny@bcherny
@EricBuess Yep, working on improving clarity here to make it more explicit
English
Starting tomorrow at 12pm PT, Claude subscriptions will no longer cover usage on third-party tools like OpenClaw.
You can still use these tools with your Claude login via extra usage bundles (now available at a discount), or with a Claude API key.
English
I'm seeing a lot of bad takes on this way of working. You really have to try it. Annotations are the way of the future, it erases a lot of friction. There are a few people who have been working like this for several months (inspired by @aidenybai's github.com/aidenybai/reac…).
Internally, at our startup we have an element annotation tool that lets us make a bunch of changes to the design at once, live on the codebase. You write down all the things you want to change, copy that as a prompt and paste it into your agent. It saves a LOOOT of time. It makes so much easier to use the codebase as the source of truth for design, no need to maintain Figma files.
erik@flowstated
cursor now has design mode (⇧+⌘+D) - click to edit, drag to draw - shift + drag to box things in - add directly to chat with ⌥+click
English
Rafael Mendiola retweetledi
@steveruizok I'm not saying that we as humans will be doing the code review
English
@steveruizok Hmm, not sure. Code review is a problem that scales. Code review at the scale we're currently working at will probably be solved, but we'll be working with massive architectures at a bigger scale at that point
English
I think code review is a passing problem and will be practically solved in six months by having smarter models and gradually eroded standards
English
@kentcdodds "So many of the people in the React ecosystem are controlling"
- Kent C. Dodds
English
So many of the people in the React ecosystem are controlling inputs unnecessarily. I know it's not my fault because I only talk about it once and it doesn't even get its own exercise. It's a portion of another exercise.
hooks.epicreact.dev/exercise/01/02…
Ryan Florence@ryanflorence
Controlled components make an imperative API declarative, but does it really buy you enough for the theoretical and implementation complexity? <input type="checkbox" value={checked} /> onClick={() => setChecked(true)} vs onClick={() => ref.checked = true}
English
Imagine if Google Chrome charged you for reactivating a tab that you haven't touched in 15 minutes
English
@GroundControl yeah the weekly limits thing is brutal. i have been doing the same rotation — whichever one isnt rate limiting me gets the next task. codex has been solid for stuff that doesnt need a ton of back and forth
English
As I'm going through Claude Code withdrawal, I ran out of my Codex weekly limit (20 dollar plan) and I'm falling back on Gemini. My favorite feature of Gemini cli is how it doesn't let you scroll up and see what it said
English
It's a mess, and I feel pretty let down. From what I gather, and I could be wrong, they increased the token limit to 1M for everyone. But their context cache only last 5 minutes. That means that if you step away from Claude Code for 5 minutes, you get a context cache miss and a huge amount of tokens get used up against your limit:
x.com/GroundControl/…
English
@captainsafia But they turned on 1M token context on for everyone, while their server context cache is 5 mins. So stepping away for a drink and coming back to resume session burns way more extra tokens now.
English
People are dunking on Anthropic for blaming the user here but I totally hear where they are coming from.
This technology is new and people are bound to mash the elevator button thinking it'll come faster when it comes to aligning the model used with the problem solved.
Lydia Hallie ✨@lydiahallie
Thank you to everyone who spent time sending us feedback and reports. We've investigated and we're sorry this has been a bad experience. Here's what we found:
English
@captainsafia A lot of the problems seem to come from the out of the box defaults
English
@pablostanley @victoria_framer That's because the real game is being all three at the same time, and the fact that they're separate roles is weird and it's actually an inneficiency
English
I feel that lately... with AI
I'm becoming less of a designer
and more of a combo of a PM and an engineer
and I don't know how to feel about it
it's cool, I guess?
English
@luongnv89 @_brian_johnson @lydiahallie @trq212 @bcherny Yeah it sucks. No one should have to think about the cost of stepping away from their computer for 5 minutes. Thank you for posting the tools you used, I'm gonna have to be monitoring the cache actively
English
@GroundControl @_brian_johnson @lydiahallie @trq212 @bcherny yeah, that is an expensive "bathroom" round trip
we can play with `cache keepalive` solution, but not so sure in aspect of tos.
English
Thank you to everyone who spent time sending us feedback and reports. We've investigated and we're sorry this has been a bad experience.
Here's what we found:
Lydia Hallie ✨@lydiahallie
We're aware people are hitting usage limits in Claude Code way faster than expected. Actively investigating, will share more when we have an update!
English
@_brian_johnson @lydiahallie @trq212 @bcherny From another reply, I guess the issue is their cache timeout. So basically if you're working on a claude code session, go to the bathroom and come back, you lose the cache and you get charged for the entire conversation again
x.com/i/status/20396…
Luong NGUYEN@luongnv89
English
@GroundControl @lydiahallie @trq212 @bcherny yeah if idle time is chewing through usage that's brutal. i'd be annoyed too
English