June 2026

One of the first things I learned in Machine Learning was that reducing loss is important. After all, a lower loss usually means better predictions. So naturally, I assumed that if a model keeps reducing the training loss, it must be getting better. But that assumption isn’t always true. Sometimes a model becomes so good at fitting the training data that it starts memorizing it instead of learning meaningful patterns. As a result, it performs extremely well on […]

Ver mais

Why memory is the missing layer of the agent stack, the seven problems any real memory system has to solve, and the shape of an architecture that survives them. Part 1 of 2 on building memory for AI agents. This part is the “why.” Part 2 is the architecture. ~10 min read. On Monday you tell your coding agent that your team uses pnpm, not npm. It apologizes, fixes the command, and carries on. On Thursday it runs npm install again. […]

Ver mais

The two LangChain middleware layers that strip secrets, compress memory, and keep agents coherent across two-hour sessions Last week, you gave your AI agent an API key. The agent used it, completed the task, and moved on. You closed the tab. What you did not do is tell it to forget. That key is still there. Sitting in the conversation history. Every model call the agent makes from that point forward is carrying your secret as context automatically, silently, with no error […]

Ver mais

A year ago, every developer I worked with was learning to write prompts. Six months ago, the same people were learning to design agents… Continue reading on Towards AI »

Ver mais