AI & Agents
Toolkits and agents for building AI features and shipping code.
Two lenses here: the toolkit I use to build AI features into products, and the agent I use to write the code itself.
Overview
When I add AI to a product I want a typed, streaming-first toolkit — not raw fetch calls to a provider. And the way I actually ship code now is agentic: this very site is being built with an agent in the terminal. Both belong in the same place because they're the same shift — AI as a first-class part of the stack and the workflow.
Tools
Vercel AI SDK
TypeScript toolkit for LLM apps — streaming, tool calls, structured output. Already in this repo's stack; my default for AI features.
Claude Code
Agentic coding in the terminal — how this very site is being built. My default for non-trivial implementation work.
My pick
For building features, the Vercel AI SDK — it's the typed layer the AI Elements components are built for, so the data and the UI line up. For writing the code, Claude Code is how I work day to day; the Pipeline section shows where it fits in my workflow.
This page is about AI as a product/dev tool. For reusable agent assets — Skills, MCP servers and Prompts I drop into agents — see the AI section. Different lens: those are configuration for the agent, not products I integrate.