How many of us started trying to hydrate everything in Pokopia? Before we learned about the wonders of rain π
07.03.2026 12:40
π 1
π 0
π¬ 0
π 0
How many of us started trying to hydrate everything in Pokopia? Before we learned about the wonders of rain π
Literally off the charts
The moat is trust, in this case for long-term maintenance and qualities not captured by testing. Same for much else that AI is impacting, such as hiring, journalism, etc.
Generating masses of code youβll never look at is not a relevant facet of AI, at least today.
Codex: "No further changes were made after your message. Your snapshot work is still Ω ΨΩΩΨΈ on this detached HEAD at e4f0854, and trunk was not modified."
It's what sorry?
I'm playing about with vibecoding something I'd never have the energy to write otherwise, which is a personalised blend of Raycast Notes and Antinote.
It's a lot of fun and surprisingly effective once the agent has tests. I do feel like I've become a weird product + QA hybrid, though. It's tiring.
Tried playing with ChatGPTβs personalisation settings and wow did I not expect to get so viscerally irritated by its tone. Switched it right back.
The daily struggle to get the Sonos volume slider to land on a neat increment of 5. It's somehow always n + 1.
I don't follow. Like -> unlike -> like -> unlike. They're sequential and the first like hasn't completed yet. Why couldn't I safely fully cancel the first unlike and second like before their actions have even run?
In the "Cancelling queued Actions" scenario, it'd be nice to have guidance where it's not a mutation that's safe to cancel in flight, but your actions cancel each other out so you want to abort queued actions. A boolean like button is a very common example of this.
A bit specific, but has anyone else noticed their Nix x Direnv \$PATH order is wrong in @zed.dev? It's causing stuff in /usr/bin, like outdated git and curl, to take priority. External terminal has no such issue.
How hard is migrating PDS? And whatβs the exit story if hypothetically npmx goes down?
Viewed through this lens it's now much cheaper to let the code represent whatever the business logic currently is, be that e.g. five distinct variants, one common variant and one distinct, or one common variant. The cost of switching between these scenarios used to be a limiting factor.
As I've become more experienced I've trended towards embracing duplication on the frontend, anticipating product churn.
I'm wondering if LLMs change the calculus on this; there's now a much smaller time cost associated with duplicating something which has outgrown its abstraction or vice versa.
A LLM/agent just tried to use the ?? JS operator against an Option object (fp-ts/Effect). And the overarching logic of the entire block was completely incoherent anyway.
On the other hand, the feedback loop with agents able to use documented commands (typechecking, tests, etc) is extremely cool.
I suggest trying squash -i -t next, so you neednβt move around history at all. Then from there megamergeβ¦
Eagerly awaiting support for non-linear branches π
To be fair with type driven development it always felt a little like this. Presumably test driven too.
SelfCI is super interesting: app.radicle.xyz/nodes/radicl...
Essentially: βworks on my machineβ is largely a solved problem, our machines are faster than rented dev servers, most of us work with code from people we trust, and debugging remote CI sucks.
Quick update on this: I might be able to use an onKeyDown prop albeit on a different component in its hierarchy. π
The dialog's ref (Base UI) isn't ready on first render. It also feels like this goes against the spirit of "you might not need an effect" now that we have ref callbacks. Unless I've misunderstood?
Agree. You can lower the volume under accessibility in the earbuds/headphones settings. I like 25%.
I want to set up a keydown event for keyboard nav, however my use case is inside a dialog where I need to attach the event to a rendered element via ref callback (instead of e.g. document.body in useEffect). I don't want to re-run the ref callback, but the event handler needs access to fresh state.
Can I super duper really not call useEffectEvent inside a ref callback? I have what feels like a natural use case and couldnβt find alternative guidance.
Top tier reference
β¨Jujutsuβ¨
Shells also struggle with things like Jujutsu revsets. IMO we might be overdue for a reimagined shell.
I would be. *gestures wildly across the Atlantic*
Sure would be nice to have something like arkregex / @arktype.io for the new URLPattern API... π
(Nobody else for now, each invite carries tree risk π )
Sure thing, DM me :)