this is the future of team collaboration humans and agents, working side by side fast, focused, and beautifully minimal introducing async: @asyncfeed

Just past midnight… so why *not* ship something new? Introducing: @WorldClassAI 🏆 AI changes fast. New models drop daily. It's impossible to keep up. Which ones are actually good? Which ones will still be good tomorrow? With a focus on video, audio and 3D, World-Class AI cuts through the chaos: ⚡️ Compare real outputs side-by-side ⚡️ Save time and money testing models ⚡️ In-depth reviews, filters and rankings ⚡️ Find underrated gems before they blow up It's early, but I couldn't wait any longer. Let's go.

Rapid-fire launch sequence is underway Goal: release 20 creations by the end of June A wide range spanning apps, art, music, videos and games The first release is live today on @Spotify 20. @theprompttheory 19. 18. 17. 16. 15. 14. 13. 12. 11. 10. 9. 8. 7. 6. 5. 4. 3. 2. 1.

A decade ago, I started working on what would become SpecUI, a spec-driven code generator aimed at revolutionizing the way we build frontend and backend applications. Originally named "ZappJS", the vision was ambitious and groundbreaking, but the first version was weighed down by its own complexity. It was buggy, slow, and struggled to scale. Then, six years ago, I shared a big release in an article (see below), full of excitement and hope. But it didn't take long for the cracks to show. Maintaining specs and generators turned out to be far more challenging than I had anticipated. The experience just wasn't easy, and the project was hard to work with. Since then, I've been on a relentless journey to simplify everything. I've learned a lot along the way. For instance, I used to think it was a good idea to offer a collection of generators (like React, Webpack, npm, etc.) for developers to choose from. It sounded cool in theory, but it added a whole layer of dependencies to manage. Now, SpecUI uses a single generator for each spec, making dependency management for spec engineers virtually nonexistent. I also initially believed having multiple specs (one for each generator) would be beneficial. My reasoning was that if the spec schema changed, it would lower the blast radius by isolating each generator. But it ended up being a headache. Variables had to be reused across specs (like the app name), and there just weren't that many shared parts to justify the complexity. Consolidating everything into a single spec has made things far simpler and more intuitive. Another big lesson was about the editor. I tried hard to build my own custom editor, and while it was cool, it was ultimately a dead end. Existing editors are already incredibly powerful, and developers still want the flexibility to write code, not just 100% specs. Extending capabilities within existing editors is a far better approach than trying to reinvent the wheel, and it saves a lot of time and effort. Integration with manual code was another game-changer. My original approach allowed developers to inject custom code within specs, which made everything hard to maintain and tied it to the custom editor I built. Now, SpecUI generates code from YAML that seamlessly connects with manual code. For example, you can define a React component in YAML, and if you want to add a custom hook, you just tell the spec to use it. And finally, I stumbled upon how enjoyable and effective it is to write code templates in TypeScript using string interpolation instead of using template engines like Handlebars. It's still compatible with Handlebars of course, but for me, using TypeScript makes templates more readable and robust, with built-in types that reduce the chance of breaking things. After years of refining and simplifying, I’m proud to say SpecUI has finally arrived where I wanted it to be. It’s intuitive, fast, and designed to scale beautifully. And next week, I'm rolling out a new feature that makes building custom code generators dead simple. The past decade has also been shaped by the rise of incredible new tools and technologies, making SpecUI's development more efficient and robust. The ecosystem has caught up, and now the vision is fully realized. medium.com/@zappjs/wtf-is-za…

Happy Next.js Conf Day

Shipping soon: Information Overload (IO) Your AI isn't dumb - it just needs better info Especially for dev tools. Instead of hunting down updated docs, imagine a free resource that’s already done the work. Using IO, you can copy & paste AI-optimized docs in one click Watch how I copied @xAI API docs, released 2 days ago, and pasted them into @v0 to generate code for Grok-2 Coming soon to a browser near you for @biomejs, @nextjs, @shadcn, @FastAPI, @EffectTS_, @htmx_org and many more

Let's cook together

I'm building the OpenAPI / Swagger specification for UIs Goals: 1️⃣ Standardize UI component libraries 2️⃣ Interoperability across frameworks 3️⃣ Built-in accessibility and validation 4️⃣ Support web and native platforms 5️⃣ Doc + test + code generation What should it be called?

It's live! SpecUI is my life's work. Today, after nearly a decade, it gets a major workflow upgrade: a VS @code extension ◆ Code generation on every keystroke ◆ Spec generation, powered by @Vercel AI SDK ◆ Integrated with @nextjs, @shadcn, @framer motion, @TauriApps and more

Now imagine this—for frameworks, languages, component libraries, API providers, test tools, monitoring & logging, animation libraries, containerization, CMS providers, bundlers, system architectures, documentation, accessibility, design patterns, CI/CD pipelines, vendors, perf...

experimenting with ai + specs

The Spec Engineer - Specializes in writing specs to build entire applications - Writes API/UI specs more than code - Automates implementation with code generation

8 yrs ago today, I shared how to generate code from specs using @zappjs Today, I'm still advocating for spec-driven development and wearing black 🖤 In the AI codegen era, SDD cuts latency & improves output. AI outperforms in speccing vs coding, focusing on logic over syntax.

Prism is getting better at training and fine-tuning other AI models. It's even making AI better at puns 😆 All of these user prompts and assistant completions were generated by Prism's AI model:

Kicked off a cool project yesterday 💥 Prism: easily and quickly fine-tune AI models. My favorite part is using another AI model to generate a custom dataset for the fine-tuned AI 🤯 Been itching to work on this for a bit - and after @Vercel announced new AI integrations last week, I thought it was a good excuse to work on this and play around with the Vercel AI SDK on something new. I also wanted to spend more time with RSC and server actions, and honestly, it's been awesome. Blown away by what I was able to whip up in just a handful of hours. So far, this works with @OpenAI, and I'm adding @Replicate + other models/providers that support fine-tuning. Excited to see this come to life! Check out the (very early) live demo: prismeditor.vercel.app

const [x, y, z] = gen

hear me out

A few months ago, I used @TauriApps to create a desktop app to watch @Vercel deployments from my menu bar. Now my menu bar is full of apps. I have one-click access to the things I use most. Here's how I did it: github.com/zappjs/tauri-mba-…

Finally: my first non-work-related push of 2024. But instead of shipping new features, I've spotlighted some existing ones. See more → zappjs.com

How many auth providers is too many?