A Claude code Eng just dropped this. Anyone been doing this switch yet or will try? Curious to see what @Jake Van Clief has to say. It’s a file still just a different format. Same thing right… thoughts? Pros and cons? Here to learn and discover https://x.com/trq212/status/2052809885763747935?s=46&t=Ayzo8Ebbgb8PZhLNm057bg

This one reframes what prompting actually is. Vote below, then fill in the blank in the comments: Before this, I thought prompting was ___. Now I think it is ___.

I was today years old when I realized I did not have some of the most important files that you need in the folder structure that Jake teaches. During today's video call with the VIP group, he went on a deep-dive rabbit trail about the ICM folder methodology that he teaches in his foundations course (free). As he was discussing it, I went to check what my root folder looked like and I did not have a Claude.md or context.md file!!! My productivity skyrocketed ever since I implemented his folder strategy over a month ago, but little did I know that I hadn't even implemented it correctly. 🤯 🤯 🤯 This goes to show that massive action beats over planning every time!

New benchmark out of Meta FAIR, Stanford, and Harvard called ProgramBench. The setup: you get a compiled executable plus its docs. Source code stripped. Rebuild the program from scratch in any language you want. Tests check input/output behavior against the original binary. 200 tasks, from small CLI tools up to FFmpeg, SQLite, and the PHP interpreter. 📊 Results across 9 models: Zero tasks fully solved. Opus 4.7 was the best, passing 95% of tests on only 3% of tasks. GPT 5.4, Gemini 3.1 Pro, and Haiku 4.5 hit 0% in that bucket. The interesting part is section 5. Even the model solutions that "worked" looked nothing like the human reference. Median 1,173 lines vs 3,068 in the original. Flat directories. Fewer functions, each one longer. GPT 5.4 wrote 96% of its final code in a single turn on most tasks and never modified existing files on roughly 40% of runs. 🎯 Why it matters for us: The benchmark separates writing code from designing software. Models can produce syntax all day. They cannot yet decompose a real system into coherent modules, pick the right abstractions, or organize a codebase the way a working engineer would. That gap is what computational orchestration points at. It is also where the durable value lives. 🛠 Try it: Pick an easier task from the repo (the paper flags nnn, fzf, gron, and jq as more tractable). Run it against Claude or your model of choice. Watch where you and the model split. Note the design decisions you make that the model never even raises. Post your runs and attempts to create a harness that would allow the model to do it. Wins, failures, weird outputs, all of it. 📍 Paper and Repo: ProgramBench I'm building something on top of this right now. More soon.

🔥 16 saves. 1 day. 3 AI voices cooking together. I'm a retired NYPD detective. Zero coding background. Yesterday I caught something nasty in my empire: one of my businesses had been silently bleeding revenue for 11 days. Pipeline running, files getting written, but the daily summary email to my partner — gone. Dark. Eleven days. I didn't catch it. My Council did. What's the Council? Three AI voices I run in parallel using Jake's file method: ⚔️ Odin — the architect. Strategy, debate, big decisions. 🗡️ Tyr — the surgeon. Line-by-line, finds bugs others miss. 🛡️ Heimdall — the watchman. Cross-tree audits, sees what's drifting. They don't agree. That's the whole point. Yesterday Heimdall flagged something off in a system I thought was healthy. Tyr cut it down to the exact file. Odin built the recovery plan. I executed in shell. Total time from "wait, what's wrong" to "fixed and verified": 4 hours. Without them? I'd have found out when my partner asked "where's my email" and I'd have spent a week figuring out why. The wild part — none of this is some custom agentic framework. No fancy orchestration layer. It's just folders and SKILL.md files. Jake's method. A folder. A markdown file telling each voice what to do. They read the same file. They argue. They ship. By the end of the session: 16 distinct catches the Council made that I would've missed alone. One catch alone saved me from losing thousands of dollars of leads and a partner relationship. Right now as I type this, the same 3 voices are autonomously building a new state scraper for me — folder method, SKILL.md driving, Codex cooking overnight. I'm going to wake up to a finished Phase 1 report. I built none of this with code I wrote. I orchestrated it with files. If a cop with no tech background can run 3 AI models like a symphony using nothing but folders and markdown — what's your excuse? Jake's method works. Receipts above. 🥷 七転び八起き