The Agentic World #2

The Agentic World #2: Agentic Memory

Giving agents memory is easy. Making that memory trustworthy is harder.

About

AI agents are getting memory. They can remember conversations, learn from previous interactions, retain preferences, and use past experiences to make future decisions.

But remembering something and remembering it correctly are two very different things.

What happens when an agent remembers something that is no longer true? When two memories contradict each other? When an incorrect memory influences every decision that follows?

As agents move from completing individual tasks to operating autonomously over weeks and months, memory becomes critical infrastructure. The challenge isn't simply storing more information. It's building systems that know what to remember, what to retrieve, what to update, and when to forget.

The Agentic World #2 is about building agent memory that actually works. We'll explore the architectures behind persistent memory, the challenges of maintaining accuracy over time, and what it takes to make memory reliable enough for production agents.

Because an agent is only as trustworthy as the memory it relies on.

On Stage

Keynote 1: Using Knowledge Graphs as a Reliable Memory Base for Agents
Speaker: Aditya Shrivastava, AI Researcher@HydraDB

Storing and representing data in the right format is important for agents to be able to retrieve them for ambiguous queries. Here we will explore how representing data using knowledge graphs can help answer complex queries.


Keynote 2: Beyond the limitations of traditional memory
Speaker: Rayyan Zahid, Founder@ImmersiveCommons

An incorrect answer affects one interaction. An incorrect memory can affect every interaction that follows.

As agents accumulate knowledge, they face a new set of challenges: outdated information, contradictory facts, memory poisoning, and the accumulation of irrelevant context.

How do we prevent agents from building their future decisions on incorrect information? Can they recognize when something they remember is no longer true? And how should memory systems decide what deserves to be preserved, updated, or forgotten?

Fireside: The Memory Problem — What Should Agents Remember, and What Should They Forget?

Speakers:
Aditya Shrivastava, AI Reasercher@HydraDB
Rayyan Zahid, Founder@ImmersiveCommons
TBA

The next generation of AI agents won't just execute tasks. They'll accumulate experiences, remember previous decisions, and continuously adapt their behavior.

But persistent memory introduces difficult questions.

Should agents remember everything? Who decides which memories are important? What happens when memories contradict each other? Can agents learn from their mistakes without reinforcing incorrect assumptions?

And as memory becomes increasingly valuable, should it belong to the user, the application, or the infrastructure powering the agent?

We'll bring together builders working on agent memory to discuss where current approaches fall short, what reliable long-term memory requires, and whether agents can ever truly know when to forget.

Schedule

• 6:00 – 6:15 · Doors and introductions

• 6:15 – 6:45 · Building Reliable Memory for AI Agents

• 6:45 – 7:10 · When Agents Remember Wrong

• 7:10 – 7:40 · Fireside: The Memory Problem

• 7:40 – 8:00 · Drinks and networking

Who It's For

Founders, AI researchers, engineers, and anyone building agents that need to remember, learn, and operate reliably over time.

Whether you're building memory infrastructure, developing autonomous agents, working on knowledge graphs and retrieval systems, or figuring out how to move agents from demos into production, this room is for you.

Details

📅 Thursday, 2026
🕕 6:00 – 8:30 PM
📍 San Francisco
🎟️ Free. Seats are limited.

Come for the understanding and adoption of right kind of memory architecture. Stay for the questions we haven't figured out how to answer.