elderlydoofus

Just released v0.2.0 of agents-svelte, my slop port of the @cloudflare agents sdk React APIs. 🎉 github.com/joshuadavidtho… @threepointone @whoiskatrin

@dreamsofcode_io Same for me yesterday! (Something with the upstream provider? Coincidence? Who knows??) What’s frustrating is putting processes in place to mitigate this (good agents file, curated skills, decentish well organized/factored code, using a good harness, etc) and it not mattering.

Last night was probably one of the worst experiences I’ve had with LLMs / Agents generating code (GPT-5.5 btw) Every commit produced regressions, the agent was constantly taking shortcuts, and there was enough hallucinations to make one develop a new sense of reality. I came incredibly close to rage quitting using Agents all together. It’s times like those I wonder if we’re all being stupid letting these things loose just for some multiples of perceived productivity.

@dprophecyguy @kitlangton x.com/elderlydoofus/…

@kitlangton no one has beaten me.

Oh wow. I have a potty mouth.

@thdxr Makes sense to me, it’s called Mother’s Day where you celebrate their motherhood so double down, grab three or four neighborhood kids, the more the merrier, here ya go

tomorrow's liz's first mothers day im gonna make myself scarce so she can fully enjoy all aspects of being a mother

@badlogicgames This makes me feel marginally better about my stats

amateur

I have a clanker rage issue I guess

@badlogicgames In that same vein, I adapted @htmx_org grug brained dev site to an agent skill. Maybe it saves tokens, who cares.. it does seem to knock a little sense into the clankers, on top of it tickles me to see them stick to the grug bit. github.com/joshuadavidtho…

TIL about "caveman mode" to "save tokens". how many tokens in a session are actually model output? i think i'll become a gardener.

@davis7 Yeah, but they still love overusing $effect and shaping things like React. All solvable with good prompts or skills but it ain’t perfect.

Hottest take I do not have data for (yet, I'll figure out a way to measure this at some point): agents are better at Svelte than React at this point This has been true for me personally and if u just think about how each of them work it makes sense with how much simpler svelte is BUT THE TOP DOWN RENDERING A TALL LANKY MAN WITH DARK HAIR AND A CONSPICUOUSLY MISSING MUSTACHE CRIES 1) skill issues stop using stupid models (gpt-5.5 on your machine, it's better on mine trust) 2) since svelte is compiled and the runes look/feel like react hooks it's much harder to foot gun yourself then in something like solid 3) there are a lot of really good built in ways to debug this stuff like `$inspect` It's in a very good state right now - remote functions built in (TRPC) - runes are killer - really good docs + setups for agents - an amazing global state/context system - easy to build/deploy anywhere - no virtual dom - all the svelte specific libraries we need - every single normal JS SDK just works right out of the box with svelte. No special hooks or libraries needed.

Natural evolution of a good engineer, trying to wrap a non-deterministic tool with a deterministic system. Gets you down to the bare metal and you really get a feel for how the models work and behave. Plus you are beholden to the tool changing system prompts of from underneath you. Go for it!

I’m gearing up to build my own agent orchestration system. Are we all doing this now?? What stage of grief is this?

What a milestone, my first GitHub sponsor! Thanks for making my week @markjaquith 😄

I thought this would played out by now, Twitter doing the speed run it always does, but this rules. More of this, show how weird these little aliens in a datacenter are. Though I’m glad they try to prompt it away because it would get old. But right now it’s delightful.

We’re announcing: VibeBench, a new benchmark for what actually matters — how models feel when used on real work by experienced software engineers. But, we need your help. Here’s how it works: 1. An initial cohort of 1000 qualified software engineers (join: vibebench.standardagents.ai) 2. Groups of 250 evaluate new models for 2 days on real work. 3. Participants subjectively rank the model relative to other models they have experience with. 4. On day 4 a report is released with objective results derived from the subjective tests. How can you help: 1. We all need this benchmark to exist, but for it to become reality, we need an initial cohort of 1000 qualified software engineers. If that’s you, please join! vibebench.standardagents.ai 2. Repost this! We need to reach as many qualified engineers as we can find. 3. Share this initiative with everyone on your engineering teams. Together we can make this benchmark a reality for all of us.

@badlogicgames @dillon_mulroy Excellent, I can finally delete my slop extension adding it as a provider and start using the official builtin slop provider

People of pi.dev Cloudflare Workers AI release. We want to see forking company and triangle company battle it out to the death. So since we had Vercel Gateway in pi for quite a while, we now have merged a PR that adds CF Workers AI as well. CF AI Gateway yet to come.

Sucks to hear and I feel that. I hope things start looking up soon. I go in waves, where a few weeks I am on a roll vibing with the new agentic approach and knocking things out, and then a few weeks where I miss the old way and hate AI and the labs (usually around the time Dario says some stupid shit about "we're destroying all jobs, just warning you! sorry! our hands our tied, we can't help it! give us all your money please though, but don't use our models in a way we don't approve"). It's exhausting.

I’m going through the craziest burnout I’ve experienced in my ~17 year career I’ve been sick for 16 days now, haven’t even been able to go for walks I kind of fucking hate AI I think all of these things are related

@simonw Actually now that I think about it a partridge in a pear tree svg might be a good step up benchmark for these models

@simonw 🎵 and a partridge in a pear tree 🎵

This week's edition of my email newsletter features 4 pelicans riding bicycles, 1 possum on an e-scooter, up to 5 raccoons with ham radios hiding in crowds, 5 blog posts, 8 links, 3 quotes and a new chapter of my Agentic Engineering Patterns guide simonw.substack.com/p/gpt-55-chatg…

@jessfraz You know you can just say stupid, works just as well and doesn’t make you look like a fucking asshole

The best thing Apple did over the last 10+ years was move off Intel to their own chips, and you all are shocked the hardware chief got CEO, don’t be retarded.

@badlogicgames To be clear, the code still sucks shit and has no good taste, but it at least usually works in the way you want it to.

@badlogicgames I really wish the GPT models were better to talk to, bounce ideas off of, iterate on plans.. Because they are all like talking to a freshly painted beige brick wall, no vibes. They can code circles around the Claude models but they got no chill.

None of mine are what I would consider cool, but they've been mildly useful (just choosing the smaller single file ones here): skill-requires-path: gate a skill from loading based on a path existing github.com/joshuadavidtho… read: overwrite the built-in tool to allow for "reading" a directory (`ls`-ing it) with a hint github.com/joshuadavidtho… rg-replace-warning: detect `rg -r` and warn that it's replace not recursive (so, SO many models get confused by this, it's so annoying) github.com/joshuadavidtho…

What's the smallest (but comparatively coolest) extension you built for pi?