Already introduced myself into the Discord a while back, so this is more of an update. I'm in Discord more now because I'm running Lyceum and Clief Notes in parallel, along with Eduba, and the Discord is where the useful stuff shows up first. Case in point: Ari's reasoning scaffolds post. I put all seven, plus the 8-item Sound Test, into the standing rules my Claude reads at the start of every session. The one I had nothing for was pixel quarantine. My setup would look at a screenshot and tell me what it saw with full confidence. Now it has to measure the image with a script before it claims anything. I also built a leak-scrub list for my client work: names, base IDs and file paths it has to grep for before anything goes public.

I'm in The Vault now and working my way through it, its amazing! Right now I'm building folder systems for trade businesses. I start each one with the same question: what is the pain, and how can I solve it? The first one is a painter who works for builders. His pain is the purchase orders he gets paid on. They are often wrong, and nobody checks them, so I'm building a tool that checks each PO against his price list and what he actually measured on site. What drew me here was the Workflow Audit Package. I'm pointing it at my own office first. On Monday, the person who does our accounts payable is going to run the interview on their invoice entry. I also liked the Harbor Studio example in the Workflow Templates, because it shows you what a good result looks like before you build anything. Over the next month, I want to take what the AP interview finds and start automating whatever part of our invoice entry is worth it. I also want to get the PO checker running on one of the painter's real jobs. If anyone has automated part of AP, or run discovery with a small business owner, I'd like to hear how you approached it.

Something I read in The Age of AI made me rethink how I use ICM. My natural tendency is to approach ICM through business process orchestration. I define the stages, describe the steps, and try to make the whole workflow explicit. But sometimes I take this too far and turn the stages into strict rules. The result becomes over-engineered, uses more tokens, and often works less naturally with AI. I recently had the opposite experience. We replaced a detailed process with a simple skill that described the essential pattern, and the results were much better. That made something click for me. Maybe stages should act as an interpretable scaffold, rather than a fixed algorithm. They can clarify the purpose, context, constraints, handoffs, and expected outcome, while still allowing the AI to work out the best path. AI may free us from having to define an entire process in advance. But if we specify less, evaluating the outcome becomes even more important. We need to check the quality of the result, capture what we learn, and use that learning to improve the scaffold over time. Does anyone else start with ICM and then gradually slip back into rigid, rule-based process design? How do you decide what needs to be specified and what should be left for the AI to infer?

Hey, I'm 72 hours into the school community. I've been learning so much. The course has been amazing. I've already met some awesome people in here. I just want to give a general thank-you and some positive feedback saying: - Great product - Great community - Love what's going on - Learning a lot - So much value here - Appreciate everyone involved Keep cranking!

What is Jev? A newly released decision model: you give it text and a set of options, and it tells you which option fits and how sure it is. And it does so extremely fast and cheaply. I did several tests. This is what I saw. In my own test it labelled 14,007 emails in 4 minutes for about $0.50. Running the same job through Sonnet would have cost about $42 and taken a lot longer. On a smaller run the answers were as good while being 80 times cheaper and fast enough you could use it practically in real time. Why does any of this matter? Let me explain in an example of a workflow that would crank on this: You have incoming email. Send it through a decision tree: - Is the email for support or finance? - Jev - If support, what product is it for? - Jev - If for a TV, is it about the recall, or which of 3 TVs? - Jev - If recall, send canned message - code - Depending on which TV, assign to X person - code - Search documents based on the question and draft a first response - LLM The expensive LLM no longer does all the decision tree work it would have in the past. Jev does that for about 1/80th of the cost. The LLM (Or Human) is at the end of the tree not spending time pointing it in the right direction. The recall is an example of adding a new or temporary category. With Jev that means writing one more sentence describing it. No retraining, no labelled examples. If it's a recall, send them where to go for it and no LLM is called at all. It's cheaper than it looks. Questions about the same email can all go in one call: support or finance, which product, recall or not. The whole top of the tree comes back in one round trip. Only the LLM step at the bottom has to wait. Where it needs care There will always be ambiguity in text. LLMs and Jev are susceptible to this. Jev returns a confidence score that you can choose how to interpret. Is 70% good enough to move to sending a response before a human looks at it? It's cheap enough to run on every email even when a human still reads them all. It sorts the pile and flags the unsure ones first.