In this episode of Humans of AI, I sit down with my friend Eric Zhu to unpack what are the frameworks and architectures powering modern AI products today.

From our times building Semantic Kernel, Autogen, and GraphRAG at Microsoft, we offer a retrospective on the evolution of this space and where we think it'll go in 2026 and beyond.

We dive into how AI systems are actually being built today, why early agent frameworks were brittle, and what’s changing as models, tooling, and abstractions improve. Eric shares insights from hands-on experience building AI systems, including lessons around memory, orchestration, open source, and why building has become dramatically faster than even a few years ago.

We also talk about the human side of all this—how builders should think about careers in AI, avoiding burnout, and navigating a fast-moving field without getting overwhelmed.

Whether you’re an AI engineer, founder, researcher, or just curious about where things are headed, this conversation goes deep into the why behind modern AI systems.

Find the podcast on:

YouTube: https://youtu.be/9FqByZJrTN8

Spotify: https://creators.spotify.com/pod/profile/alex-chao/episodes/...

Apple Podcasts: https://podcasts.apple.com/au/podcast/the-hard-truth-about-a...

How are you building AI agents in 2026? Are you still using a framework?

The memory insight here is underappreciated. Once you cross the line from stateless chatbots to stateful agents, memory becomes infrastructure — not a feature you bolt on later. We learned this the hard way: agents that can't reason over what happened yesterday make the same mistakes on loop. The real shift in 2026 isn't better models, it's better state management. Persistent memory, proper context windows that survive across sessions, and failure recovery that doesn't require re-prompting from scratch. That's where the actual engineering challenge lives now.