So today’s addition to my Slack command centre is the beginning of my content creation process. (Thanks to @David Vogel for suggesting Buzz yesterday, sadly it doesn’t accommodate buttons yet, but I’m monitoring developments) I’m working on my Pat’s Film Club workspace, and yesterday: - Created the process to pull all the image assets of a film that I’m going to show from Movie Database, to create marketing materials. - Created the process where I could officially program the film once I’ve got a confirmation from the cinema about the booking. This then creates an event in the Community calendar, and a space for the Film in the current screenings discussion area, pre-populated with topics for the trailer, Film details and an essay on why I chose the film (which was empty). - Wired these processes to buttons in a Slack app so I could run them at will without having to remember slash commands or have a model interpret what I want. Today was about capturing my thoughts about the film to create a master resource in my own voice that could then be repurposed for all materials - website copy, posts, email sequences and social media captions. Using the steps I’d already been doing in my Second Brain system, I adapted them to run through ICM folders. I used @Jake Van Clief ’s ICM Architect skill, but was also inspired by @Ry Mac ’s 3-jobs-rule. The process: - I identify the film - I talk for as long as I want with all the details I want to say about the film, my story with it, etc - It keeps the raw version, then fact checks the details and presents me with its suggested corrections for my approval - It creates a tidied version just using my words, and presents it for my approval - It then takes the approved version and uses it to see if there are any new insights to add to my “Pat’s voice” reference file, and suggests them to me (if any) - If I approve them, it updates the voice file and then has the raw and tidied versions of my content, ready for repurposing. - Wired this process into my Slack app - where everything is done.

This weekend I will be mentoring a youth group (15-20 years old) on the biggest university in South America, USP. We're doing a hackaton to create in 3 days the seeds of a Comic Book that will explore the benefits and perils of AI to the workforce and life in general. I was invited to serve as a mentor on AI related issues. I will bring the results next week!! Have a great weekend, High Tea and Lyceum meeting!!!

I understand how the CLAUDE.md works for context and routing, and that part is becoming much clearer as I build more of these systems. The bit I’m still struggling with is what to do with large reference material — things like guidance documents, API documentation, product specifications, research papers, manuals, or even an entire book — where you want Claude to be able to find the relevant information when needed without having to load huge amounts of material into context. To give a practical example, I’m currently building an ICM system for medical educators who supervise paediatric trainees. The idea is that a supervisor could use it to help with things such as: - preparing for supervision meetings; - answering questions about training requirements and guidance; - keeping track of meetings, deadlines and important dates; - completing the required supervision and progression forms; - maintaining appropriate information about individual trainees; - and helping with forms or reports relating to other trainees they are asked to assess. - It’s still very much a work in progress, but I think the basic architecture is coming together nicely. I’ll share the CLAUDE.md as well so people can see what I’m trying to build. The recurring problem I’ve found — not just with this project, but with most of the folders I’m building — is how to deal with the reference library behind the system. For example, this medical education system may eventually need access to fairly substantial training guidance, curricula, policies, forms and supporting documentation. I obviously don’t want all of that sitting in CLAUDE.md, nor do I want Claude reading hundreds of pages every time somebody asks a relatively simple question. What I’d really like is something closer to: Question → router → identify the relevant reference source → retrieve only the relevant section → answer using that material. In other words, keeping the ICM folder as the structure and source of context, while having a larger reference library that is searched only when required.

Finally giving my view on JEV in my classic history and fundamentals fashion but... This one is a little different. It's my first time making a fully animated, pixel art video, and I'd honestly love to know what you think of the style. Should I make more like this? I made an ICM that fully automates most of the process i was able to get this video built in about three or four hours. Talk about a fun animation style right? Might redo a bunch of the videos in the classroom with this.

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.