Vol. I · No. 579Rochester, New York, Thursday, September 24, 2026Generative UI · Page B11
All workSection B · Page B11

Generative UITools4U

A Canvas of Widgets You Order in Plain English

An infinite canvas where an AI agent turns a plain-English request into a live widget wired to your own data.

Graphic5 steps
  1. DetectChecks for missing connectors before generating
  2. ResearchAgent pass with search, extract and weather tools, up to 5 steps
  3. SpecStructured pass returns a Zod-validated ToolSpec
  4. RenderNative primitives or a Sandpack React sandbox
  5. RefreshTrigger.dev reruns scheduled tools
What happens inside Tools4U, in order.Graphic: The Sahil Bachu

Tools4U is a canvas for tools you make by describing them. Ask for "a card that summarizes my unread emails" and an agent builds that card, drops it into a free slot on an infinite grid and can refresh it on a schedule. Everything you make lives side by side, like a desk covered in sticky notes that actually do things.

Generation runs in two passes with the Vercel AI SDK. The first pass is agentic: the model can search the web with Tavily or Exa, pull pages with Firecrawl or check the weather, for up to 5 steps. The second pass is strict. It has to return a ToolSpec that passes a Zod schema, covering what kind of tool it is, its layout and where its data comes from. Splitting it this way lets the first call wander and keeps the second call honest.

Splitting it this way lets the first call wander and keeps the second call honest.

A ToolSpec renders one of two ways. Simple tools use 8 native primitives I wrote, like MetricCard, ListFeed and ChartBlock. Anything custom runs as React inside a Sandpack sandbox. Before generating, a small pre-pass checks whether the tool needs an integration you haven't connected and asks you to connect it first through Nango OAuth. Edits work the same way: the model gets the existing spec and changes it in place.

Underneath sit Supabase auth, Postgres with row-level security and Realtime, a Trigger.dev job every 5 minutes that refreshes scheduled tools, and your own Anthropic or OpenRouter key, encrypted on the server. The canvas is React Flow, with drag, resize, recolor and skeleton cards while a tool is being made.

It builds clean, but I haven't run it end to end with real keys yet, and per-provider data shaping for the connectors isn't written. It overlaps with SecondScreen on purpose. Both ask what happens when software is something you describe, and this one answers on a big screen instead of a phone.

More from The Work

See all 25

Browser Agents

The Job Hunt Agent That Refuses to Hit Submit

JobOps finds jobs, tailors a resume for each one and pre-fills the application, then stops and waits for me.

A job-search command center that runs on my laptop. It pulls postings from five sources, scores them with headless Claude, tailors a LaTeX resume per job and pre-fills applications in Chrome. It never submits. I review and send every one.

Continued on Page B9

Agentic AI

Your Coding Agent Fills the Gaps in Your Resume

A local-first orchestrator that reads a student's resume, proposes projects to fill the gaps and has their own Claude Code or Codex build and deploy them.

ResumeGod reads your resume, suggests projects that would fill its gaps, then drives your own Claude Code or Codex to build and deploy them. It finishes with a personal site and a fresh LaTeX resume. Nothing is hosted centrally.

Continued on Page B15

Multi-agent

Hire AI Employees, Then Walk Over to Their Desks

A pixel-art office where every AI agent is an employee with its own desk, skills, scheduled jobs and Docker container.

PixelHQ turns the open-source SkyOffice into a single-player office for AI agents. Walk up to an employee, chat with it, give it skills and scheduled jobs, and feed a shared knowledge base. Each employee runs in its own container.

Continued on Page B17