Building Agents in JavaScript
summary
A JavaScript agent is built that retrieves vectorized Wikipedia snippets, composes a five‑fact email about the Olympics, and sends it successfully via an automated demo using llama-index for orchestration, a vector DB for semantic search and resend as the email tool. The talk argues structured output (example: Zod schemas), function calling and short/long memory are essential, shows a working demo, notes doing it raw takes ~100 lines and manual parsing while an orchestrator makes it reliable and safer.
AI-written summary · about
chapters (10)
- 0:00Building agents in JavaScript
- 1:05Why JavaScript developers care about AI agents
- 2:52LLM and agent history
- 14:55Key components of an AI agent
- 20:37What it takes to run an agent
- 27:44Crafting system prompts
- 29:10Live demo: retrieval agent
- 43:07Defining tools and importance
- 51:29Risks and guardrails
- 55:46Resources and closing
related talks
24mYour First JavaScript AI Agent
19mBuilding Agents at Cloud Scale
01h 00mHow to work with generative AI in JavaScript
43mThe AgentVerse - Build fast, Scale smarter
45mAI agents in practice - beyond the prototype
41mBuilding LLM Agents – Challenges and Learnings
25mFrom Prompt to Production: Building Autonomous AI Agents in Node.js
30mBuilding Robust, Secure LLM and Agentic AI Workflows