Sabitlenmiş Tweet
Matthew Shilts
1.2K posts
Matthew Shilts
@mfshilts
I love software, building businesses, and farming in my spare time. -- CTO @ Solera Network
San Diego, CA Katılım Aralık 2010
661 Takip Edilen150 Takipçiler
No question, in my experience strongly typed languages generate far less slop than weakly typed languages when you are letting AI do the code generation. And if you have some basic patterns and interfaces for it to start out with, then its even better.
English
First review of our movie! bloody-disgusting.com/movie/3950043/…
English
My movie: Big Buck Massacre finally dropped. Its been years "in the can" waiting for release. It was an incredible week of filming with friends. Horror comedy is the genre. Its "so bad its good":
youtu.be/PxrfzEPoRrg?si…
YouTube
English
So, I built one for my game system. Its intended to be a re-usable NPM package. I'd love some feedback. github.com/mshilts/incarn…
English
For small apps or games we need a better auth system. Something that works for AI agents. I want a system where I don't need to depend on one of the SaaS auth providers, and I don't want to have to store passwords and usernames. So, the answer is Public Keys and SSH.
English
@nikillinit Generally I agree with this, but one addition which I believe is critical around universal basic care, which includes wellness to keep people healthy. The definition of “basic” is key. For more info: github.com/mshilts/afford…
English
IMO most of US healthcare would be fixed* if we did a few simple things
1) divorce employers from providing health insurance
2) Use a price setting mechanism that was transparent (government setting prices or price transparency)
3) Everyone in one risk pool (e.g. on the exchanges, single payer, etc)
*fixed in the sense that you would be making tradeoffs between cost, quality, and access to care - but at least those tradeoffs would clearer
English
@jonallie This thread has some excellent insights. I need to rethink some of my modern assumptions.
English
Matthew Shilts retweetledi
Half-formed shower thought: we need a better version control paradigm for AI assisted coding.
The agent speedup feels great on greenfield/solo projects, but when you start collaborating with other devs, things can fall apart. Either you all exchange huge PRs (abandoning code review, and dealing with gnarly merge conflicts), or stick to a traditional flow with reasonably sized changes, in which case you spend a lot of time waiting for changes to land.
Old style version control systems had the notion of files being "checked out" by a user. This was painful, but was a pretty good signal that someone was making changes to a set of files that might affect what you wanted to do. Obviously that had huge downsides, but I wonder if that "signal" doesn't still have value. In theory you could get something like this from git, with automatic / frequent push and pulls. You'd have to deal with the messy commit history, but...maybe the existing "functional chunk commits" strategy is an anachronism
English
@bindureddy Our chief engineer @CherylFoil did this early last year and we saw a 10x productivity lift for our engineers. This year is glorious. But, not everyone has a GOAT engineer to pull it off without AI. So blessed to have a rockstar team.
English
Pro Tip - If possible abandon legacy code and re-do the entire app on a full stack AI-native platform
Constrained AI with guardrails is MAGICAL
English
@steipete we need a new codex plan in between $20 and $200 for OpenClaw. There is a sweet spot somewhere in there that is a huge missed opportunity. Given your new influence, maybe you can help?
English
@BenjaminBadejo How did you get it, it’s own signal account? I did exactly same for Google workspace, works amazingly well.
English
Once you give OpenClaw its own dedicated Google Workspace account (with read-only access to your own email and calendar, plus giving yourself manage/edit access to its account), give it a dedicated phone number, get it to use vision / browser tools with mouse and keyboard controls for full navigation abilities, and have it download every document production program it can think of, the possibilities are endless.
It made me an accounting system, calendar booking system on a website, and invoice production / editing / delivery / archiving system.
In less than 15 minutes. Whenever it needed my input, it would just message me on Signal and I would respond there, and then it would continue its work.
Amazing.
English
Matthew Shilts retweetledi
**🦞 Fix: OpenClaw Google Chat "not responding" / 401 errors**
If your OpenClaw Google Chat channel shows stopped in openclaw channels status --probe and keeps auto-restarting, you're hitting a bug where the webhook channel crash-loops and creates duplicate handler registrations, causing 401 ambiguous webhook target.
**Step 1: Config** — Make sure your audience config matches what Google actually sends:
bash
openclaw config set channels.googlechat.audienceType app-url
openclaw config set channels.googlechat.audience "YOUR-TAILSCALE-HOST.ts.net/googlechat"
**Step 2: Patch the crash loop** — Edit extensions/googlechat/src/channel.ts in your OpenClaw install (find it with find $(npm root -g)/openclaw -path "*/googlechat/src/channel.ts").
Find the startAccount function. After the startGoogleChatMonitor() call, add these lines before the return () => {:
typescript
// Webhook channels register a route and return; keep the promise pending
// so the gateway framework does not treat this as "channel exited".
await new Promise((resolve) => {
ctx.abortSignal.addEventListener("abort", () => resolve(), { once: true });
});
**Step 3: Restart**
bash
openclaw gateway restart
Verify with openclaw channels status --probe — should show running (not stopped).
⚠️ The patch gets overwritten on openclaw update — re-apply after updating until this is fixed upstream.
English
Matthew Shilts retweetledi
Matthew Shilts retweetledi
This is the power user Claude Code workflow
From the creator himself
TL;DR:
• Run 5 Claudes in parallel, numbered tabs 1-5, system notifications when input needed
• Use Opus 4.5 for everything. Slower but requires less steering, so faster overall
• Start in Plan mode (shift+tab twice), iterate until good, then switch to auto-accept for 1-shot execution
• Slash commands for “inner loop” workflows you repeat daily. Lives in .claude/commands/
• Single CLAUDE.md per repo, checked into git. Team contributes. Every mistake becomes a rule
• Tag @.claude on coworker PRs to add to CLAUDE.md as part of review
• Subagents for common PR workflows: code-simplifier, verify-app, build-validator, code-architect
• PostToolUse hook auto-formats code. Handles the last 10% before CI
• /permissions to pre-allow safe bash commands. Skip –dangerously-skip-permissions
• Claude uses all your tools via MCP: Slack, BigQuery, Sentry, etc.
• Give Claude verification. Browser tests, bash commands, test suites. 2-3x quality improvement
• Long tasks: background agent verification, agent Stop hook, or ralph-wiggum plugin
Boris Cherny@bcherny
I'm Boris and I created Claude Code. Lots of people have asked how I use Claude Code, so I wanted to show off my setup a bit. My setup might be surprisingly vanilla! Claude Code works great out of the box, so I personally don't customize it much. There is no one correct way to use Claude Code: we intentionally build it in a way that you can use it, customize it, and hack it however you like. Each person on the Claude Code team uses it very differently. So, here goes.
English