Liz Zhang

Talks & videos

FEATURED TALK

Memory Engineering for Long-Running AI Agents

Agent Memory Night, SF Tech Week · Oct 8, 2026
AWS Builder Loft, San Francisco

Why agents get worse as they run longer, and how a graph fixes accuracy, latency, privacy, and cost. The talk uses my own open benchmark: 20M tokens vs 4.4M for the same 200-message conversation.

Run the benchmark ↗

Talks, demos, and podcast

INVITE LIZ TO SPEAK →

GraphRAG vs Vector RAG for AI Agents (Live Demo)

Hack for Humanity

Every Personal Agent Promises Memory. Here's What Breaks

FalkorDB, host

Other stages

  • Liz Zhang: Patch the Conference Hackathon: opening talk

    Patch the Conference Hackathon: opening talk

  • Liz Zhang: Agent Opus Video Hackathon: host

    Agent Opus Video Hackathon: host

  • Liz Zhang: Talk at the EIT event in Hungary

    Talk at the EIT event in Hungary

Tutorials and demos

AI Agent Memory System Design at Scale in 13 Minutes

FalkorDB

How to Build a Voice Agent w/ Pipecat (Full Code)

OpenLine, part 2

I Built A Voice Agent that Calls Leads in Seconds

OpenLine, part 1

AI Startup Product Demo

As CTO

Your Old Prompts Make Fable 5.1 Worse (6 Fixes)

This AI Agent Saves Me Hours Every Week

Shorts

AI Agent Memory System Design at Scale

Who Owns Your Personal AI Agent's Memory?

Same AI agent, 4.5x fewer tokens. I changed one thing.

An Anthropic engineer told me only ONE of these scales

I Attended an AI Conference Inside a Museum of Dead Tech in Silicon Valley