Daniel Strongin

Expo + React Native is why a non-coder can ship a real mobile app. One project becomes an iPhone app, an Android app, and a website. An AI agent edits a file → your phone updates instantly. Package it into a real installable app when you're ready. The prompt I used (paste into Claude Code or any coding agent in an empty folder): "Set up an Expo + React Native app in this folder. Install everything, then start the dev server so I can scan the QR with Expo Go on my phone." 10 minutes later I was building on my phone.

@karpathy Awesome! Can't wait to see what you learn with Anthropic trickle down to your educational material

Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.

@trq212 Oh yeah, definitely have to do this. Yeah, there always are choices that are made for you that you don’t know about

a prompt I've been using a lot recently: implement <SPEC> and while you do, keep a running implementation-notes.html file (or markdown) with decisions you had to make weren't in the spec, things you had to change, tradeoffs you had to make or anything else I should know

I’m getting more convinced the useful agent pattern is not “let it run everything.” It’s: 1. agent writes the fix 2. tests prove the contract 3. repo log records the decision 4. external actions wait behind a human gate Today: 28 tests, 3 issues closed, 6 automation pages shipped cleanly.

@trq212 I use HTML for everything now. You opened up pandoras box lol

HTML continues to be undefeated

@TheReviewRewind Thanks! 😊

@Strongin_ Nice! Have fun on your honeymoon!

Using Claude Code to plan my honeymoon. France, June 4-23. It's booked: 4 stays, 5 trains, 2 wine tours, a Michelin dinner, a Seine dinner cruise, Eiffel Tower top floor, Louvre, Musée d'Orsay, a food tour. Budget in markdown, EUR auto-converts to USD. Every confirmation PDF lives in the repo.

@TheReviewRewind @mattpocockuk It for sure helps a ton of actually implementing systems that work

@Strongin_ @mattpocockuk This! I need to be using this to help me create workflows my business

I’ve been using @mattpocockuk’s agent workflow ideas: grill the design first, turn it into a PRD, split it into issues, then TDD each slice. It turned duplicated automation into a shared runner that’s easier for AI agents to navigate, test, debug, and change.

@mattpocockuk What about one skill orchestrating a bunch of small skills to complete a complex task

Long skills are such a red flag to me - Hard to audit (and therefore, trust) - Hard to edit (more text, harder to maintain) - Expensive to run (more text, more tokens) The shorter the skill, the better IMO

Codex computer use is pretty freaking good. It checked Telegram/Teams, found a song on Spotify, looked at email stats, and scanned messages/emails for anything needing attention. All from one concise prompt. Starting to feel genuinely useful. Using Computer Use to post this.

Built a public system-bug-investigator skill for coding agents. It keeps bug fixes from turning into symptom patches: identify the failure surface, trace related files/services, check blast radius, choose the safest first fix, then verify. github.com/danman231/agen…

Listening to @salimismail on @PeterDiamandis’ Moonshots clarified something for me: Most AI agent failures aren’t prompt problems. They’re architecture problems. So I made a public skill for designing/reviewing agent workflows before they get messy. ➡️ github.com/danman231/agen…

6/6 Install it: npx skills@latest add danman231/agent-skills --skill agent-architecture Repo: github.com/danman231/agen…

5/6 So I made a public skill for this: agent-architecture It helps an agent design or review an AI workflow before it turns into a vague, overpowered, hard-to-debug automation.

1/6 Most AI agent failures are architecture problems, not prompt problems. The model can be smart and still fail if the workflow has unclear roles, too much context, no approval gates, and no validation.

@perplexity_ai Why would isn’t it on pro subscriptions?

GPT-5.5 is now available on Perplexity for Max subscribers. GPT-5.5 is also rolling out as the default orchestration model in Computer for both Pro and Max subscribers.