Angelina YangFounder and Host, TwoSetAI · AI builder for visibility and discoverability

"We Killed RAG, MCP, and Agentic Loops" | Paul Iusztin, Author of LLM Engineer's Handbook

16 May 2026 · 1 min

Watch on YouTube

0:52

Paul Iusztin argues that RAG and Memory-Centric Prompting hit hard limits: context windows max out at 64,000 tokens and repeated memory queries add latency. For controlled applications, skipping these techniques often works better. AI-generated code gets treated like any other compilation step, though it still needs extensive testing before deployment.

Chapters

in our particular use case. Everyone was using MCP everywhere. If you have hammer, you see nails everywhere. Load all the data into the context window, it will just sum up to 64,000 tokens at maximum. The model, if it doesn't find the answer, it will start to query the memory again and again. Zigzag adds extra latency. If I use Cloud Code and I just want to hook to my Gmail, I still think it's amazing. But when you build an application where you have full control over, I would just skip it. Usually the The answer was more in simplicity, but it's hard to find this simplicity.
What do you think of bytecode?
Python is compiled many, many times until it reaches bytecode. Do we read the bytecode to make sure it's clean bytecode? No, we just trust it. AI-generated code is just a compilation step. I was super anxious. I still am. I need 30 minutes of preparation.
I need to treat to coffee in Portugal.
I love discussing, but I need 3 days of recovery.