Today I will show what we are doing with agents in @seed.hyper.media come and join us! luma.com/xauvgrx6

4 booleans = 16 possible combinations. Only 7 are valid. The other 9 are bugs waiting to happen. Why I model UI workflows as state machines: seed.horacioh.com/writing/butt...

A rich text editor is a document system with a schema, transactions, selections, and identity — not a decorated <textarea>. That's why I built Seed's editor on ProseMirror. Full writeup: seed.horacioh.com/writing/rich...

AI made code cheap but didn't make knowing what to build any easier. The bottleneck was never typing — it was understanding, modeling, and product sense. Expanded my thoughts: seed.horacioh.com/writing/ai-e...

A trust graph for OSS shouldn't be owned by any single platform — not GitHub, GitLab, or anyone else. Maintainer reputation shouldn't evaporate when you change forges. The graph has to be portable.

The big question: how does OSS change when plausible code is nearly free? My answer: contribution friction should scale with trust distance and change criticality. Full proposal here: seed.horacioh.com/writing/grad...

The MVP I want to see: scoped expiring vouches, contribution policies as signed docs, hop-distance calculation over the graph, and a GitHub Action surfacing trust distance in PRs.

What's the right primitive for OSS trust? Not stars, followers, tenure, or signed commits alone. I keep coming back to scoped, revocable, reputation-bearing vouches.

AI did not just make code cheaper. It made maintainer attention more expensive. In OSS, the valuable contribution is not only "here is code." It is: the problem, the reproduction, the tradeoff, and the smallest useful fix. Good contributions compress maintainer attention.

A good OSS trust model has to be portable. Every forge should read the same graph. Maintainer reputation shouldn't disappear when you move from GitHub to Codeberg.

Open source ran for 30 years on a hidden assumption: a working patch = real human attention. AI broke that. Plausible-looking code is nearly free to generate, but human understanding is not.

Betting on XState for frontend logic is never a bad call. useState works at first — but logic always gets more complex. When it does, not having a statechart means untangling implicit state manually. Start with XState early. You'll thank yourself later.

Local-first is not just offline mode. It changes the product relationship: your work starts with you, survives the network, and syncs when the system catches up. That one shift changes architecture, UX, recovery paths, and trust.

Rich-text editors are not textareas with bold buttons. They are document systems: schemas, selections, history, identity, comments, embeds, collaboration, import/export, and recovery paths. The toolbar is the easy part. The product system underneath is where the complexity lives.

Most complex UI bugs are invalid states wearing a nice design system. Once you add loading, retry, cancel, partial success, background sync, and recovery paths, you are not just managing component state anymore. You are modeling product behavior.

A good statechart is a product spec developers can run. It captures what can happen, what cannot happen, and what the UI should do next. That is the real value for complex product flows: ambiguity becomes visible before it turns into bugs.

Most complex UI bugs are invalid states wearing a nice design system. Once you add loading, retry, cancel, partial success, background sync, and recovery paths, you are not just managing component state anymore. You are modeling product behavior.

creating content strategy pipelines using AI is fun. How y'all are doing it?

Open source shouldn't become closed. A typo fix from a newcomer should be easy. But a parser change in critical infra shouldn't get the same treatment as surface-level edits. Friction should scale with trust distance and change criticality.

OSS security asks "is this signed by the right identity?" The harder question: "should this person be trusted to touch this code?" 2FA and signed commits help with identity — they don't answer trust.

Anti-pattern: storing Promises in React state. Promises are reference-stable — no re-renders. Track the status (useState<boolean>, useMutation), not the work itself.

Slop PRs are not just annoying — they are a denial-of-service attack on maintainer attention. A trust system should keep low-signal contributions away from the front of the review queue without punishing real contributors.

The key trick in onMouseLeave: setHoveredBlockId(prev => prev === blockId ? null : prev). Only clear if YOU are still the hovered block — prevents race conditions. Works for: block editors, table rows, sidebar navs, any one-active UI.

Sometimes the best optimization is just better state management 🧠 What patterns do you use for this? 👇

From N useState hooks to one shared context. Only one block highlights at a time ✅

Stop giving every component its own hover state 🛑 N components × N useState hooks = multiple blocks highlighted at once. The fix: lift hover to a shared context. One source of truth. 👇

AI didn't just make code cheaper. It made maintainer attention more expensive. That's the real open source security crisis — and most of the tooling we're building doesn't address it.

Im starting to create this weekly summaries of my work with the help of AI and I kinda like it seed.horacioh.com/projects/see... What do you think?

clawdbot + n8n on my old linux machine at home...

We adopted Tamagui to unify our UI across web, desktop & mobile. Ended up needing 96GB RAM just to run dev 😵‍💫 We're now migrating to Tailwind + Shadcn — and seeing real performance gains. Full breakdown of the journey ↓ seedteamtalks.hyper.media/updates/migr...

Just finished a blazing round of SudoRush! Can you beat my score? mode: Medium ⚡️⚡️ score: 14 accuracy: 93% points: 19 👉 sudorush.app #SudoRush

Just like painters once feared photography would steal their jobs, people now worry AI will replace workers. History shows new tech changes jobs, but rarely erases creativity or the human touch… #AI #creativity

We are building REAL freedom from the start!

I rewatch this today and It bugs me how some frontend developers don't see how much value this brings to UI development. hhg.link/short-posts/...

In @seed.hyper.media, decentralization is not negotiable.

Knowledge communities will only operate in full capacity under decentralized circumstances.

I fucking love @atproto.com because it is mostly decentralized Today I’m thinking about how to integrate it into the (unstable) decentralized network at the heart of our publishing product

I'll be first to admit that the @seed.hyper.media protocol is poorly formalized and documented Although we'd love to improve docs and stability of the Hypermedia protocol, this has been a strategic choice Our focus is on building a great product that everyone can use, and the protocol will follow!

Exciting to see growth in ATproto and I’m sorry if this comes off as rude… How is this “unbannable”? It is centralized just like bluesky, and it’s not open source(??) Apps are only distributed with Apple and Google, not yet globally released Seems easy to ban. Already banned, if you are in Europe

You can do it now with a @seed.hyper.media website. You can see it on hhg.link

This week I updated OBS and it was not opening on my machine 😭 I promise to fix it next week and stream again! I have a lot of new features to show 😬

The web community needs to renew our focus on building decentralized trust, community preservation, and accessibility of publishing for non-technical people This is why I love working on @seed.hyper.media every day Our product is open in every way, and you are all invited to join the cause!

SR-71 Blackbird hhg.link/short-posts/...

This was a fun watch! hhg.link/short-posts/... kudos to @veritasium.bsky.social

I was a believer that AI was evil. and now I can't think of working in tech without it. Now I'm able to create side-projects, websites and test my simple ideas in a matter of hours!! this weekend I made 2 websites for family in less than 3 hours... this is insane

Streaming again! twitch.tv/horacioh Today we will be doing some State machines UI work :)

FIXED! (it was an easy fix tho 😁)

I just decided to start streaming today! twitch.tv/horacioh come and hang out!

Hello, world! We are Seed Hypermedia, and our mission is to preserve trust and collaboration on the open web.

Shipping updates shouldn't be a headache! This guide breaks down how to set up a custom auto-update system for your Electron app—no third-party services required! 🔗 Check it out: seed.hyper.media/blog/how-to-...