CodeHagen

Introduction HagenKit - the next OSS boilerplate ✨ And its free for everyone 🤯 hagenkit.com Launch your new project in 2 minutes – from clicking the button to live preview ✨ Techstack: 🌐 App: @nextjs ☁️ Hosting: @vercel 🎨 UI: @shadcn 🗂️ ORM: @prisma 🛢️ DB: @prisma ✉️ Mail: @resend 🔒 Auth: @better_auth I wanted to make it as easy that i can for you to get started with your new project. → We make the Github-repo → We make the Database → We push this to Vercel. Hope you like it ✨

@shadcn @TheSpacerr Love it, keep then coming 🔥

@TheSpacerr @CodeHagen Will continue working on chat interfaces for a bit. Bring all these together.

@shadcn /typeset is stupidly good. Imported one CSS file, wrapped 3 surfaces in a typeset class, done. Instant editorial typography ✅

@shadcn Lets go 🔥👌

Introducing shadcn/typeset. You know how you render markdown and get back plain, unstyled HTML? Headings, paragraphs, lists, tables. So you style the elements one by one: font sizes, line heights, spacing. You do it for your blog. Then you do it again for docs. Then again for the chat. Every time, you're fighting the same thing: sizing and spacing. To fix this, we created typeset.css: one file that styles everything inside a typeset container. It lives in your project, so you can change it directly when you need to. And we made it work beautifully with streaming markdown.

@hagaetc Yes 🔥

Haaland!!! 🐐🇳🇴🚣❤️‍🔥

@TomasJansson @shadcn It was basically for me to see the UI at first and tweak. Then i make a design.md for it so that the agent can look at after. Have so far worked like a charm ✨

@CodeHagen @shadcn Won’t the page take up a very large part of the context if you feed it to the llm? Wouldn’t it be more effective to let the AI figure out what kind of UI element it needs and then provide a way to find that element in your design system?

This made me develop MVPs x10 times faster 🤯 I create a /playbook page at the start of every project now. Just init a project from @shadcn and your ready to go This page lists all the UI components my project will use: ◆ Page shells ◆ Layout patterns ◆ Component structure ◆ Design rules Give the AI this context upfront and it stays consistent across the whole build Can basically design the full MVP in one go now.

@RhysSullivan This is good - Great job

@runsonai @herdrdev Thanks @elie2222 have been talking good about @herdrdev - Maybe i should try

@CodeHagen I recently moved to @herdrdev to have opus/fable be main orchestrator, but then it can manage claude/codex/pi all together. Then plugged my hermes with it as well to be able to continue what's going on.

What am i missing? 👀 I think my multi-agent setup works well, but maybe I'm not tokenmaxing 🤔 Running 5 AI agents in parallel right now: ◆ Fable as orchestrator ◆ Opus 4.8 / Sonnet 5 as sub-agents ◆ Each agent handles a different task Is this the right way to do multi-agent workflows or is there a better approach? What are you using?

@plainionist This was a DD-rom migration for a M&A product. Did i use to much? 😅

@CodeHagen What was the task that you need that much AI power?

@noahlh Thanks for the tip - Then i keep going with ✨

You’ve got it right. This is the current meta. The only thing I’m still torn over is whether Opus or Codex 5.5 is better at implementation. I think it’s task-dependent - anything visual/design related goes to Opus, anything mechanical goes to Codex. I honestly wish we didn’t have to think about this and could just use Fable for everything, but such is life with a budget.

@TJae @shadcn Gonna do it, just need to make the repo ✨

@CodeHagen @shadcn Open source it if you’re the man!

@truffle Thanks, gonna fix a skull for ny workflow i think.

That is my pretty close. Some tips. One fable orchestrator per complex goal. If a goal isn't complex don't use fable at all or have fable do it without an orchestrator. Let fable break down the work to sub-agents as it chooses, prompt it to consider task complexity. Reinforce that the orchestrator is not allowed to do work as context is at a premium (even if it isn't) You can throw all this in a /orchestrate skill

@frontrun @shadcn I can make a template repo this week and send it over ✨

@CodeHagen @shadcn looks great, do u have a template or prompt for this playbook?

@robinebers Do you just write a .md plan from Fable then make GPT-5.5 to follow the plan from the .md?

the most exciting combo for me is probably Fable 5 + GLM 5.2 surprised nobody is talking about it a value combo like no other

@becomevocal Yes, starting with Fable planning and use workflow with Opus usually. Actaully never used /new - Have just been using /context-save from @garrytan when i need a new session. Thanks for the tip 🔥

@CodeHagen Did you start out with fable planning? I like to start with a fresh context often, so always plan into a .md Tell sub agents to commit updates as you go along Then you can /new every few turns while retaining state Depending on your plan IDK why you'd use sonnet at all

@elie2222 @vercel Love it ✨

Introducing Atom Eve! An open-source registry of agents, built with eve or Flue. It's like shadcn or skills .sh, but for agents. You get the agent code to fully customize to your needs. It supports both eve and Flue framework. Two new agent frameworks by the @vercel and @astrodotbuild teams, respectively. Eve bills itself as Next.js, but for agents. And Flue is a similar idea, structured a bit differently and less tied to Vercel. These agents are designed to handle real tasks to manage your business. One example: SEO Improver: 1. It connects to your Google Search Console and DataForSEO 2. Understands how you're ranking and makes or suggests improvements 3. A week later it looks back at how your rankings changed and iterates again based, continuing on its work from last week. 4. Each week it continues where it left off with the latest data. (BTW if you get nothing else from this tweet, that workflow is massive alpha in and of itself.) There are lots more agents and I'd love to have people contribute. The project is open source. Give it a star on Github! Try it today: atomeve.dev Star it: github.com/elie222/atom-e…

@shadcn @base_ui Love it! Just made a new project today with Base UI actually ✨

Starting today, we are making @base_ui the default component library in shadcn/ui. First, a bit of history. When shadcn/ui launched in 2023, it was built on Radix. At the time, nothing else came close. Headless. Accessible. Composable. Fast forward a few years and the team who built Radix are building something new: Base UI.

@shadcn 👀

We are going to make one big change to shadcn/ui this week.

The Cofounder model is changing ✅ Used to think the technical cofounder was everything But now? i'd give 50% to someone who deeply knows the space Why: ◆ AI makes development easier each day ◆ the hard part is knowing what to build for the niche ◆ domain experts see opportunities others miss Execution speed matters less if you're solving the wrong problem What do you think?