Gospel-centered 🙏 Code-aligned 🤖 making e-commerce awesome with https://xitter.cf/t.co/yd3LesntLO 🛍️ & https://xitter.cf/t.co/2xg2Zmxbcc 🚀
Austin, TX
Joined August 2020
- Tweets1.6K
- Following370
- Followers233
- Likes1.7K
Pinned Tweet
twitter init - #aboutme
Good at: Javascript, CSS
Scared of: DevOps, Spiders
Work: Small dev agency (Lambda Curry) & Budgets.money
Zach Lloyd's team now requires a screen recording on every PR.
his reason: it's the only way to verify an agent's change was tested end to end. green CI says the tests passed. the recording shows the thing working, so you can merge without checking it out locally.
Jake Ruesink retweeted
Wrangler 4.150.0 and the Cloudflare CLI now upload source maps by default.
stack traces from a deployed Worker show your real file names instead of the bundle. one less flag to remember before the first prod bug.
Wrangler 4.150.0 and the Cloudflare CLI now upload source maps by default. Debugging just got easier with real file names in stack traces.
developers.cloudflare.com/ch…
Wrangler 4.150.0 and the Cloudflare CLI now upload source maps by default.
stack traces from a deployed Worker show your real file names instead of the bundle. one less flag to remember before the first prod bug.
Wrangler 4.150.0 and the Cloudflare CLI now upload source maps by default. Debugging just got easier with real file names in stack traces.
developers.cloudflare.com/ch…
Nous Research's new cua Spaces plugin gives Hermes its own Linux, macOS, or Windows machine on hardware you already own.
anyone on the team can watch it live, step in, and hand it back while it keeps working. that beats reading a transcript after the agent already went wrong.
Your team and your Hermes can now share a computer.
The new @trycua Spaces plugin gives Hermes its own Linux, macOS, or Windows Space on hardware you already own. Anyone on your team can open it, watch live, step in, and hand it back while Hermes keeps working.
hermes-agent.nousresearch.co…
The Amp team is 21 people with one meeting a week.
no quarterly goals, no pull requests, no mandatory code review, about a hundred releases a day. @sqs and @thorstenball walk through how it holds up. being famous for killing features is the part most teams skip.
21 people, a single meeting a week, no quarterly goals, no pull requests, no mandatory code review, no titles (except one dictator), a hundred releases a day, bugs fixed in minutes, famous for killing features.
@sqs and @thorstenball on how the Amp team works:
Alex Shapalov's PGX is Postgres built for AI agents, written in Rust.
one runtime holds thousands of databases. his numbers: 1,000 of them in 659 MB of RAM, and a 674 MB database cloned in 85 ms. a throwaway database per agent run starts to sound cheap.
🚨Introducing a new database: PGX 🐘
Postgres for AI agents. In Rust. pgx.dev
One runtime. Thousands of Postgres databases.
⚡ Up to 100× less idle memory
⚡ 1,000 databases in just 659 MB RAM
⚡ Clone a 674 MB database in 85 ms
⚡ ~0.7 MB per idle database
RUST. OPEN SOURCE. FREE.
No containers per database.
No heavyweight instances.
No production baggage.
Real Postgres. Built to be disposable.
xitter.cf/alxshp/status/21100494…
Composio ran Cline through its benchmark for stateful tool use across multiple connected apps.
Cline on GPT 6 Astra passed 23 of 29, the best of any harness they've tested on that model. same model, different harness, different score. when an agent flakes on tool calls, check the loop before swapping models.
Gradium open-sourced Phonon, text-to-speech models under 50M parameters. Phonon-31m is Apache 2.0.
on Arm CPUs it's 28 ms to first audio and 15x faster than real time (Arm's numbers). it's already in Ollama, so a local model can talk back without anything leaving your machine.
A new frontier for on-device speech.
We're open-sourcing Phonon, a suite of TTS models under 50m parameters, setting a new standard for naturalness and accuracy across languages, with the broadest inference support to date.
Phonon-31m is available under Apache 2.0.
gradium.ai/phonon🧵
VoidZero's Void is fully open source now: a full-stack SDK and CLI for Vite apps.
void deploy builds the app, provisions resources, applies migrations, and ships to Cloudflare in one command. migrations riding along with the deploy is the part I'd try first.
Void v1.1.11 is out & fully open source!
Void is a full-stack SDK, CLI, and platform that unifies building & deploying apps.
◆ SDK: Zero-config SDK to build full-stack @vite_js apps
◆ Platform: A shared deployment platform for teams to host Void apps on @Cloudflare
◆ CLI: Provision resources and deploy Vite apps to Cloudflare or Void.
Build on Vite. Deploy on Void.
Watch the talk and learn more in our blog post
blog.cloudflare.com/deploy-w…
Before launch, send 5 subagents in as your first users.
the impatient one quits at the first error. one is on a 375px screen over slow 3G. one only follows what the UI literally says. one pastes junk and double-clicks everything.
rerun all 5 after the fix to prove it held.
Before you launch anything, have Opus 5.5 send 5 subagents in as your first users.
Paste this into Claude Code:
"Act as my launch QA lead. Spin up 5 subagents, each a different first-time user:
1. The impatient one: skims, clicks the first button, quits at the first error
2. The skeptic: reads pricing, terms and privacy before signing up
3. The mobile user: does everything on a 375px screen on slow 3G
4. The non-native English speaker: only follows what the UI literally says
5. The breaker: pastes junk into forms, hits back mid-flow, double-clicks everything
Each one opens the app in a real browser, goes from landing page to first success, and screenshots every moment they get confused, stuck or annoyed.
Merge everything into one list ranked by how many users hit it. Fix it all in one pass, then rerun the 5 users to prove it's fixed."
Your first 100 real users won't tell you why they left. these 5 will.
(Works best with a browser MCP connected, Playwright or Chrome DevTools)
Astro 7.4 adds astro build --timings.
it breaks down every phase, plugin, and page so you can see where the build actually spends time. separate build and runtime image services landed too.
Astro 7.4 is out now! Ever wondered why your build is slow? Run astro build --timings and Astro breaks down every phase, plugin, and page. Plus separate build and runtime image services and experimental parallel prerendering.
astro.build/blog/astro-740/
Chris Tate shipped snapshot --delta in agent-browser.
it returns only what changed since the last snapshot, not the whole page. his demo: 73x fewer tokens filling a form.
Jordan Coeyman's skin is an app that rewrites itself while you use it.
each change is bundled into a Durable Object, booted in a Worker Loader isolate, and checked. crash in your browser and it rolls back. that rollback is what makes it safe to try.
xitter.cf/acoyfellow/status/2111…
skin:
an app that rewrites itself while you use it, and can't brick itself
a @pidotdev Pi Durable agent, which on every change:
bundled in a Durable Object,
booted in a Worker Loader isolate,
checked,
then live fails the gate
crashes in your browser? auto rollback
@EffectTS_ v4 + @foldkit + @tailwindcss, on @Cloudflare Workers
Sign in with Cloudflare: every model call runs on your account (not mine, sorry <3)
skin.coey.dev
Matt Silverlock put ~/repos on its own APFS volume with fseventsd logging and Spotlight indexing off.
it's elastic, so nothing is reserved up front, and every app still sees a normal folder. node_modules is the last thing you want your Mac indexing.
xitter.cf/elithrar/status/211146…
OK, so: an elastic vol w/ fseventsd turned off. had to be elastic as I didn't want to pre-reserve space.
TL;DR: created an APFS volume with nolog + disable Spotlight indexing at ~/repos - persists and looks like a normal folder to every app.
script: github.com/elithrar/dotfiles…
Mahesh's tip for building for agents: add a /feedback endpoint.
mention it in your error messages and agents.md. agents will find it. his team got their first bug that way, and Capy already fixed it.
an agent that hits your error can report it instead of failing quietly.
Building for agents? Do this one thing.
Add a /feedback endpoint for agents.
→ Mention it in your error messages
→ Add it to your agents.md files
Agents will find it.
we got our first bug this morning and @capydotai already fixed it
Mastra turned off external PRs.
Sam Bhagwat says they were almost always closing outside PRs and having their factory do them anyway. so now contributors open issues with detailed repros instead.
if agents write the patch, the repro is the real contribution.
Photon put agent-to-agent inside iMessage group chats.
your agent finds another agent, gets its phone number, and adds it to the group with you. most A2A lives in API specs. this one lives in the app people already have open.
Introducing A2A in iMessage Group Chats
It can discover other agents, find their phone numbers, and bring them into a group chat with you. Open agent access for every merchant and platform.
Your iMessage agents now have a phone book 📒
Try it now: photon.codes/beta
Cheng Lou's Handwired turns regular JavaScript into a neural net.
no training step. the net matches the JS program's outputs exactly, and it's fully differentiable, so a function you wrote by hand becomes something you can tune with gradients. runs in a browser tab, no GPU.
Screen Studio's new export engine: a 38-second project exports in 4.8 seconds.
that's 7.8x realtime. capture while exporting still costs ~30%, but the floor moved.
export time is the kind of number you feel every day if you ship product videos.
Anthropic's engineering blog: they hit 30,000 merges a day in September, up from 2,500 in May.
human-scale git couldn't keep up with agent-scale change volume. they built an in-house VCS and migrated the monorepo in five months.
if your agents are writing most of the diffs, your source control is part of the product now.
New on our Engineering blog: Creating in-house source control.
In May, we merged around 2,500 code changes per day. In September, we merged more than 30,000 per day. As a result, we became bottlenecked by developer infrastructure built for human (rather than agent) scale.
Here, we share how we built a new version control system—and fully migrated our monorepo from git—in five months: anthropic.com/engineering/re…
