Tyler Barnes, founding engineer at Mastra, introduces Observational Memory. It is a new memory system for AI agents that achieves state-of-the-art results on LongMemEval with a completely stable context window.

Unlike semantic recall (which uses RAG and invalidates prompt caching), Observational Memory compresses conversations into dense observations while maintaining a stable, fully cacheable context.

The result: 94.87% accuracy on LongMemEval with GPT-5 mini. This is the highest score recorded by any memory system to date.

In this conversation, Tyler explains how the system works, why it outperforms raw context, and how you can integrate it into your agents in under 20 minutes. We also dive into the research, the benchmarks, and what's next for Observational Memory.

🧰 RESOURCES MENTIONED

Observational Memory Launch Blog: https://mastra.ai/blog/observational-memory

📚 MASTRA RESOURCES

MASTRA?

00:00 – Intro

Shane and Abhi cover top AI stories. This week was absolutely massive! Anthropic aired Super Bowl ads mocking OpenAI's decision to put ads in ChatGPT, Opus 4.6, and GPT-5.3-Codex launched within 15 minutes of each other, and then ClawHub dropped a bombshell: 11.9% of the entire marketplace is malware.

We cover everything: Anthropic's competitive jabs, the model war benchmarks, Claude's 1M token context, OpenAI's Frontier platform, the security crisis that's reshaping how people think about agent marketplaces, Kimi K2.5's domination on OpenRouter, ElevenLabs' $500M raise at $11B valuation, and the explosion of AI video generation tools.

Plus: Perplexity's Model Council, Roblox 4D generation, Mistral's Voxtral Transcribe 2, and why Swyx finally admits evals actually help. AI Agents Hour is a weekly livestream hosted by Mastra CPO Shane Thomas and CTO Abhi Aiyer. Airing Mondays at 12PM Pacific on YouTube and X, the show covers breaking AI news, agent development techniques, and features interviews with industry experts building AI applications today.

🧰 RESOURCES MENTIONED Anthropic Super Bowl ads: https://x.com/tomwarren/status/2019039874771550516 Opus 4.6 launch: https://x.com/claudeai/status/2019467372609040752 GPT-5.3-Codex launch: https://x.com/sama/status/2019474754529321247 ClawHub malware report: https://x.com/aakashgupta/status/2019885557221061026 Kimi K2.5 #1 on OpenRouter: https://x.com/openrouterai/status/2020199664746971195 ElevenLabs $500M raise: https://x.com/elevenlabsio/status/2019048384011854087 Perplexity Model Council: https://x.com/perplexity_ai/status/2019444886114824219 📚 MASTRA RESOURCES Mastra: https://mastra.ai Learn Mastra in the world's first MCP-Based Course: https://mastra.ai/course Principles of Building AI Agents (Book): https://mastra.ai/book Patterns for Building AI Agents (New Book): https://mastra.ai/blog/patterns-book https://docs.google.com/forms/d/e/1FAIpQLSduJjc515f6RZJqtkR2ByqJZrB0iP8B7SUKnjjZE9IajH_I8w/viewform

MASTRA? Mastra is an open-source TypeScript framework designed for building and shipping AI-powered applications and agents with minimal friction. It supports the full lifecycle of agent development—from prototype to production. You can integrate it with frontend and backend stacks (e.g., React, Next.js, Node) or run agents as standalone services. If you’re a JavaScript or TypeScript developer looking to build an agentic or AI-powered product without starting from first principles, Mastra provides the scaffolding, tools, and integrations to accelerate that process.

Shane and Abhi welcome Kiet Ho, cofounder of Superset, to discuss how Superset evolved from simple WorkTree management into a full-featured tool with file editing, automation, cloud support, and multi-agent orchestration.

Shane and Abhi cover this week's top AI developments. The OpenClaw agent goes viral amid growing security concerns over its shifting identity. Moonshot AI's Kimi K2.5 ties for first place on Design Arena. Google DeepMind unveils Project Genie, advancing world model technology. Anthropic previews a new model and explores AI coding with Claude Cowork plugins. Plus, updates on developer tools, humanoid robots, and emerging AI models.

Today we discuss the Mastra 1.0 launch, we talk about the things we are learning, we discuss ClawdBot (renamed to MoltBot and now OpenClaw). We also discuss agent skills, new models, and how agents are reading more docs pages than humans.

Today we announce Mastra is officially 1.0! We talk with Kyle from ElectricSQL (and formerly Gatsby) about Durable Streams. Allie joins us to discuss the code review security problem that can be caused by generated AI code. Finally, we cover all the AI news from the last week including ads in ChatGPT, the Langfuse/Clickhouse acquisition, and the OpenAI / Cerebres partnership.

Today we discuss a rising star, we talk about the emerging agent primitives, and we cover downstream effects of a world where agents are now the primary users of a lot of products/tools. We cover Anthropic news, the Google+Apple partnership, model companies going public, AI in healthcare, and much more.

Who is Ralph Wiggum and why are developers talking about him? We also discuss the rise of opencode and if everyone can be a developer now. And of course, we cover all the latest in AI news.

Today we talk about how Abhi tried to rebuild Git over the holiday weekend. We also review everything that happened in the world of AI and agents in 2025. Finally, we give our 2026 predictions for what we think will happen next year.

Are Agent Skills going to be the new MCP? Today we do a deep dive discussion on Prompt Caching as well as compression. We cover all the latest AI news and finally have a guest from Morning Brew join the show.