Greg Brockman's Advice to AI Engineers (channeling my best Ben Lang)

🚨 UPDATE: Full Post-Mortem On Cursor Security Incident In yesterday’s thread I explained how I got drained after installing a malicious extension in Cursor. This is the deeper dive into what I found, what I did, and how you can avoid it. 🧵 👇

Claude 4.1 Opus System prompts have been leaked! Let's delve into the details of how it's constructed. The Claude 4.1 system prompt is a highly structured and modular configuration that exhibits several recurring system design patterns described in the Pattern Language for

Speed Always Wins Very nice and comprehensive new report on recent efficient architectures for LLMs.

📊🤖 Stock Research DeepAgent A stock analysis system using LangChain's DeepAgent framework to coordinate specialized agents for comprehensive market research and real-time data analysis. Check out the implementation and build your own! 🔍 github.com/sagar-n/deepag…

Model Context Protocol (MCP), clearly explained (with visuals):

The paradigm of AI subagents is going to be super interesting. There was probably some hope or belief that a universal agent would be able to handle everything you needed in a workflow by stuffing all the relevant context into the context window. But even with larger context

I built two ubertool MCPs: MCPs that have a single tool that accepts source code it evals. One for playwright and one for pexpect. I think this should be explored further. Will talk about this today at Claude Code Anonymous in London. lucumr.pocoo.org/2025/8/18/code…

This paper reveals a severe limitation of our understanding of AI. As AI becomes more capable, we begin to understand prompting better. Why? This is because prompt engineering (and context engineering) is about meta-cognition.

22 Cloud Design Patterns - You Should Know!!! 👇 📌 𝐀𝐯𝐚𝐢𝐥𝐚𝐛𝐢𝐥𝐢𝐭𝐲 𝐏𝐚𝐭𝐭𝐞𝐫𝐧𝐬 [1.] Circuit Breaker Stops a service from repeatedly trying a failing operation. [2.] Retry Enables an application to handle transient failures when it tries to connect to a service or

AI Engineering Toolkit! I have curated list of 100+ LLM libraries and frameworks for training, fine-tuning, building, evaluating, deploying, RAG, and AI Agents! 100% Open Source

Notes on PyPI's new protection against domain resurrection attacks, where an attacker registers an expired domain in order to gain access to account reset emails simonwillison.net/2025/Aug/19/py…

🔬Building Deep Research on top of DeepAgents DeepAgents sets a solid foundation, with built in planning, subagents, and file system usage It's pretty easy to build a comprehensive Deep Research on top of this! YouTube: youtu.be/geTtqyFnyHA Code: github.com/hwchase17/deep…

The AGENTS.md microsite is live! Think of it as a README for agents: a simple, open format for guiding coding agents. It’s been awesome to work alongside peers from Amp — Research Preview, Cursor, Jules, Factory, and Roo Code to bring this to life!

MoE layers can be really slow. When training our coding models Cursor, they ate up 27–53% of training time. So we completely rebuilt it at the kernel level and transitioned to MXFP8. The result: 3.5x faster MoE layer and 1.5x end-to-end training speedup. We believe our

Free book from Stanford covering all important algorithms for Artificial Intelligence. Link in comment.

AGENTS md is a simple, open format for guiding coding agents. Works like a README but designed specifically for AI Agents to understand your codebase. Single file works across Cursor, Claude, OpenAI Codex, Google Jules, and Factory AI.

DeepMind built a simple RAG technique that: - reduces hallucinations by 40% - improves answer relevancy by 50% Let's understand how to use it in RAG systems (with code):

Great AI at Meta paper. Builds a single test that shows when LLMs think too much or too little, then scores both. It targets a gap, reasoning models ramble on easy questions while fast models miss steps on hard ones. They release a benchmark called OptimalThinkingBench with

This is actually a really solid context engineering template. Kudos, Anthropic