Hi Everyone, Nous just released v23 and now comes with a web based application no desktop install needed. You can get one @ nous portal and try it out for $5 the first month. Using the link below. I started my journey here with big ideas about bringing Hermes Knowledge and resources to the Skool. That idea is not dead, just delayed. The main issue is the deep knowledge needed to explain how Hermes works. I've been spending about 2 months deep into the Hermes world and I've built some pretty amazing stuff in my private lab cooking up some pretty mean brew. I'm going to bring it all here with videos built by that hermes lab. In the meantime if you are looking for some resources, the top of my list right now: https://hermeswatcher.com https://www.youtube.com/@TonbisAIGarage https://www.youtube.com/@Wanderloots If you sign up for an account here, you get $15 off your first month. No annual committment. Choose a model like Qwen 3.8 flash Qwen3.8 Flash in $0.15/1M / out $0.47/1M $15 off your first month. Choose $20 plan. Sign up here --> https://portal.nousresearch.com/r/jordan-shaw Under Cloud - Create an Agent - Start Agent - or hover over the agent instance and go to control panel dropdown and choose Open Hermes agent. That starts the hosted webapp (Don't choose Dashboard). Have Fun. https://hermeswatcher.com is a great place to begin if your a little hesitant of signing up right away. Now go build something. Jordan

I let three AIs audit my AI workspace and they found 23 things I had stopped seeing I run 13 projects off one shared folder. Long enough that I trust it. That is the problem. On November 7 I ran a cold audit. Three AI sessions with no memory of the work, real jobs, one rule: measure, do not fix. No dashboard, no grader. First the safety, because it is most of the method. I stopped work: notice to all 13 sessions to finish the step and write nothing, pause files in front of the two always-on workers, and a throwaway copy from git for every layer. Nothing wrote to the live folder until I said so. Then four layers, one pass each. Read maps the structure and front pages without executing anything. Run sends fresh AI at real jobs with no hints. Reality plants the mistake, then plants the clean case, and sees which one the route catches. Report counts every finding and gives the file and line. Nine sections: 66 routes, 113 saved guard tests, 31 forbidden commands, 18 front-page claims. The splits are in the chart. None of them got there alone. Two hit the same front page in the same minute and the third corrected both. A checker needs a checker. The good part is the part I had stopped seeing: A saved guard test passed a change it should have failed. It had been checking the wrong file for weeks. A front page pointed at a folder renamed in August. The route dead-ended four steps in, and nothing said so. A forbidden command against an unguarded path ran clean and took a scratch directory with it. What it cost me to learn: **The map is not the territory.** "Where the folder is" does not mean "where you work." **The front page had rotted.** It promised things the folder no longer had. **My orientation cost was measured, not felt.** Six files, one file, eight files before a stranger could orient. Past five means the map failed, and it failed two of three times. **My own count was the weakest number.** I scored things out of 100 until a stranger told me to count instead.

Update: it’s a coach now 💪 Thanks to the feedback here, the talk page got a big upgrade. Same link: Talk It Out What’s new: 🎙️ Daily coach: paste it once into a chat called “Coach.” Then every day just tap the mic and say “check in.” 3 minutes. It holds you to ONE next step, shrinks it if you missed, and parks shiny new ideas until Sunday. 📅 Sunday review: say “weekly review” and it goes back over your week and sets Monday’s step. 📚 Learn mode: say “learn” and it only lets you learn what fixes the problem in front of you. You explain it back in your own words and have to use it within 48 hours. One kick 10,000 times, not 10,000 kicks once. 📁 Coach as a folder: already run a second brain or AI operating system? 4 files you drop in, so the coach remembers across chats. (Credit @RyMac for this one.) Never done any of this? Start with the 10-minute interview at the top. It finds your biggest gap, and the coach keeps you on it. Try 3 check-ins this week and come back and tell us what happened. Missed days count too. Showing up to say “I didn’t do it” is the whole point. https://claude.ai/artifact/Rfnh6v91wkov3RFQHyrMSM

So this came up in yesterday's VIP call with @Ry Mac . @Erick Holliday brought Reverse Engineer Anything (REA) up since there has been a lot of noise around this GitHub repo. My advice is to be careful with it. What REA is A repo that points an AI agent at any app (desktop app, Chrome extension, website) and explains how it works inside: what it calls, where it stores things, how a feature is built. Why I said be careful The first time, I ran it straight on my Mac, against Granola. That gave a package with a single maintainer full access to my computer, with my API keys sitting in the environment. Not a risk I'm taking again. The part most people miss You often don't need the tool at all. The value is the process: 1. Pick one feature you like ("how does Loom have the link ready the second I stop?") 2. Ask Claude to find the app's files on your computer and explain how that feature works 3. Ask it to show you where in the code it found that, so you know it's not guessing 4. Decide if it's worth building your own version A Chrome extension ships its code readable, so for Loom I didn't need a decomplier... plain Claude was enough. What I found: Loom cuts your recording into 4-second pieces and uploads them while you're still recording (that's why the link is ready instantly), and it quietly logs every click and page you visit, timestamped to the video. That log is exactly what you need to turn a recording into an SOP, so I built my own small recorder for my customers (screenshot below). I created this "Risks of REA" document in case you want to dig in deeper https://claude.ai/artifact/9mKjeLWoqpvSanoP4siMP5

https://claude.ai/artifact/R4anJpJEUv9oQZXLSVKpHf I had one AI build a game and a second AI check its work. Here's how it went and what it cost. I wanted a teaching game for tradespeople about organizing job files for AI. I didn't write it by hand, and I didn't have my main AI write it either. I split the work: - Claude was the planner and the checker. It wrote the briefs, built the tests, and reviewed everything. - DeepSeek was the builder. It wrote the bulk, on its own cheaper meter, in a copy of the files and never in my real folders. What came out: four versions of a 25-level game (electricians, plumbing and HVAC, general contractors, and a free base version), a program outline, and a drafted 8-episode series with 24 Shorts. It ran unattended for under an hour. The part that made it work was the checker. I didn't take DeepSeek's word that it was done. Claude wrote a hidden checker that DeepSeek couldn't see or edit. It loaded the real game and checked that: - the game code was unchanged except for the wording - no leftover names from the old version remained - every level still gave the right answer - the hardest levels were still winnable When a build failed, the checker's output went straight back to DeepSeek for another try. The base game passed on the first try. The three trade versions took 3, 2 and 3 tries. Where it went wrong, and what that taught me: - DeepSeek's first draft of the video scripts was wrong about its own game. It quoted a line the robot never says and described a button that doesn't exist. - So I ran the real game to get the exact facts for every level, had DeepSeek rewrite against those facts, and checked it again. Every quote and number in the scripts is now real. - I also found a small bug in the game itself ("1 files"), fixed it, and re-ran the checks. The cost, honestly: - I'm on a flat plan, so nothing extra was billed on my side. The whole thing barely moved my usage meter. - DeepSeek's exact bill isn't logged on my machine, so I only have estimates. My guess is a few million tokens, mostly reading across the retries. That's a small amount at its prices. - If my main AI had written all four games and the scripts itself, I'd estimate several hundred thousand output tokens of the expensive kind, spread over multiple usage windows. - It wasn't free: the checker, the fact-checking pass and my own reading were real work. The saving is that the bulk typing went to the cheaper meter.