JS Party: JavaScript, CSS, Web Development · Changelog Media

Building LLM agents in JS

July 18, 2024·59 min·5 clips

Tejas uses AI agents to run his podcast, discovering guests' passions and generating discussion outlines automatically.

Tejas Kumar begins with his background as a web engineer since age eight, his work leading developer relations at Vercel under Guillermo Rauch, and his current focus on generative AI at Datastax, a company specializing in retrieval-augmented generation. He distinguishes RAG from agentic workflows: in standard RAG, the developer retrieves context and injects it into the LLM prompt; in agentic RAG, the LLM itself decides when to call retrieval tools based on its own assessment of what it needs. Kumar defines agents using Andrew Ng's three-component framework: reflection, where the model generates output, evaluates it, and iterates until improvement is no longer possible; tool calling, where the model recognizes when a task exceeds its capabilities and invokes an external tool; and agent collaboration, where a capable orchestrator model directs multiple lower-capability models toward a combined goal. He also cites David Kurshe's framing of agents as implementations of the actor model from computer science, where an entity observes its environment and responds, using Pac-Man as a simple rule-based example. Kumar argues that LLM-based agents extend this pattern by replacing hard-coded rules with generative judgment. He describes his own podcast production workflow as a practical example: agents orchestrate webhooks from Riverside and cal.com to handle scheduling confirmations, episode metadata extraction, show notes generation, and distribution, tasks that would otherwise require a small team. The episode covers when JavaScript rather than Python is the right choice for agent workflows and discusses the Datastax ecosystem of tools for production RAG applications. The tone is technically grounded but accessible. The audience is JavaScript developers exploring AI integration.

As heard by us

A focused JavaScript discussion of agents, tool calls, and the tradeoffs behind making them work.

JS Party settles into a focused technical conversation with Tejas Kumar about AI and LLM agents in JavaScript. K-Ball and Tejas spend most of the time on the practical details that matter when code has to run: tool calls, how the AISDK pattern works, and why tool descriptions…

Read the full review in PlayNext →

Why you'd press play

If you want LLM agents in JavaScript without mysticism, press play.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from JS Party: JavaScript, CSS, Web Development