Yugabyte logo

Why your AI agents keep forgetting

Vector search alone won’t carry enterprise AI very far. Ajay Khanna on what agents actually need to remember, and why Yugabyte thinks the answer lives in the database.

Ajay Khanna
Guest

Ajay Khanna

Chief Marketing Officer, Yugabyte

Enterprise data marketing leader, now taking Yugabyte from distributed PostgreSQL into agent-native data infrastructure for multi-agent systems.

Follow Ajay on LinkedIn →
Anirudh Narayan

Anirudh Narayan

Co-founder & Chief Growth Officer, Lyzr

Co-founder & Chief Growth Officer, Lyzr. He built Agent Sam, the AI agent that pitched Lyzr to its Series A investors.

Follow Ani on LinkedIn →

Must watch if your agents forget everything the moment a session ends.

Chapters

Orange dot marks a key moment
Anirudh Narayan and Ajay Khanna on the How We Agent episode “AI Needs Memory,” a Lyzr x YugabyteDB conversation
Watch onYouTube Listen onSpotify Listen onApple Podcasts
Share on LinkedIn

Five moments worth pressing play for

Short on time? Pick up where you want.

  1. Moment 01

    A vector database is not a memory system

    Retrieval finds a relevant chunk. It doesn’t carry what an agent learned, or why it decided what it decided.

    A static knowledge baseKnowledge that evolves

  2. Moment 02

    Meko: shared memory at the database layer

    Agent-native infrastructure on YugabyteDB, combining four things that used to live in four separate stores (memory, knowledge, conversation history, observability) into one place.

    • Memory
    • Knowledge
    • Conversation history
    • Observability

  3. Moment 03

    Datapacks, and reasoning that travels

    The demo is where it gets concrete. Each agent keeps private memory, knowledge stays shareable, and the thinking behind a result is kept with it.

    • DatapackPortable, multi-tenant store: private memory, shared knowledge.
    • Decision tracingThe next agent inherits the reasoning, not just the output.
    • Agent actionsCalls like add knowledge, instead of raw plumbing.

  4. Moment 04

    Knowledge that compounds instead of resetting

    A knowledge base that never changes starts going stale the day it ships.

    Context should compound across agents, not end with each session.

  5. Moment 05

    The agents Yugabyte runs on itself

    Where its own teams use agents, and what open-source growth, ABM and enterprise go-to-market look like when the product is infrastructure.

Up next

All episodes

Want more of this?

Get notified when the next episode drops, read case studies or get AI agent use cases and templates right in your inbox every month.

No spam. Unsubscribe anytime.