Andrej Karpathy

A viral essay argues that every coding-agent memory plugin is the same RAG pipeline in different clothes, and that agents need a structured Markdown brain instead. We unpack the argument, the open-source Operator Memory plugin built on it, and the strongest objections from a 100-comment Hacker News thread.

Managed Deep Agents 0.8 ships a second durable memory layer scoped to the authenticated caller. Here is what user memory stores, how to enable it in memory.py, the default Slack and HTTP access rules, and how it relates to schedules.

Ponytail turns your AI coding agent into the laziest senior developer in the room: before writing code it climbs a ladder from do not build it, to reuse it, to the standard library, to a native feature. Its own agentic benchmark reports 54 percent less code and 100 percent safety. Here is how it works, what the numbers do and do not show, and how to try it without regret.

On October 2, 2026, Andrej Karpathy argued that as intelligence and code become abundant, you can ask for large, custom, discardable software artifacts, such as web apps and video explainers, that would never have made sense to build before. The shift is economic, not a new model capability.

On October 2, 2026, Andrej Karpathy posted a short ladder for understanding what language models write: controlled-language text, then diagrams, then HTML pages, then bespoke explainer videos. Here is what each rung means, the exact prompts to try, and where the advice breaks down.