Build Your First AI Agent
One central path in 90 minutes — plus six branches for where you go next
A hands-on technical workshop. Every participant ships a working AI agent on the Vercel AI SDK — our central path — and leaves with a portable Agent Card (Google A2A) that works across Claude, OpenAI, Google ADK, and no-code stacks. The point is not the framework; the point is the six primitives (model, tool, memory, loop, spec, deploy) that transfer to every agent you will ever build. Six optional branch modules then re-build the same agent on Claude, OpenAI, Google ADK, Oracle ADK, no-code (n8n/Notion/Dify), and AI-builds-AI (Claude Code) stacks.
Learning Objectives
Name the six primitives of any AI agent — model, tool, memory, loop, spec, deploy
Ship a working research-assistant agent on the Vercel AI SDK with tool use, memory, and a public URL
Publish a valid Google A2A Agent Card at /.well-known/agent.json and understand where Oracle Open Agent Specification fits
Write a 3-case evaluation suite and 2 refusal rails for your agent
Pick the right branch (Claude / OpenAI / Google ADK / no-code / AI-builds-AI / Oracle) for your next agent based on the transfer matrix
Prerequisites
- Basic JavaScript or TypeScript (Python fine for branches; core is TS)
- Laptop with Node 20+ and pnpm (or npm)
- API key for at least one LLM provider — Anthropic Claude recommended; OpenAI or Google fine
- Optional but recommended: a Vercel account (free tier is plenty) and a GitHub account
Workshop Agenda
A cold open: what actually separates an agent from a chatbot? The reveal — six primitives (model, tool, memory, loop, spec, deploy) that every agent framework is spelling differently. By minute 10, participants have the mental model that will carry every module.
Key Resources
Share this workshop
Get the Resource Pack
Receive the complete slide deck, handouts, and facilitator guide for this workshop.