Anyone curious about Decision Models?? I spent some time this week digging in to the details. Unbelievable proliferation in 3 weeks. https://opendoorlearningai.com/articles/113-decision-models-in-3-weeks/ TL;DR - A decision model takes some evidence plus closed questions and returns typed answers with probabilities. It writes no prose. A decision costs a tiny fraction of a cent and comes back in tens to hundreds of milliseconds. - One request can carry many questions. They run in parallel and independently, so nothing chains inside a call. - “113 entries” overstates the variety. 70 of the 112 open entries are built on Qwen, 18 on Gemma 4, and 20 are stock models with a decoding trick and no new weights. - On the independent Decision Index, Perplexity Decider v1.1 ranks first on its own, and Fastino GLiDE, Jev and Torchcast tie for second. Calibration, price and where you can run it separate the leaders more than accuracy does. - Measured fairly, the 27B models answer in 0.10 to 0.14 s on one local GPU, and hosted APIs timed by independent testers mostly land between 0.1 and 0.6 s. - Laya is one of the weakest open options. Every hardware tier, down to a laptop, has open models scoring 6x to 14x higher. - My proposed inbox setup: one script polls, filters, asks for four facts per email and applies rules, and the LLM runs only for mail that needs writing. Estimated $353 a month down to $28. It isn’t running yet; results will follow in a separate article.

I'm still a bit confused. What are the precise use cases? Who is the ideal customer? Missed calls -> Recovered Revenue for plumbers sounds good but are missed calls really the bottleneck in their business?

Gave our company’s first sales pitch yesterday. Full data architecture rebuild, connecting disparate systems, detailed reporting in real time, local LLM layered on top, delivery of everything so they own the systems. $115k solution Prospect didn’t even flinch… “wow this is exactly what we want” Follow up meeting with the board already confirmed and on the calendar.

https://youtu.be/9dYcwOkpCE8?si=xR-v5G1P27L7KPIU Fun - dropped yesterday. Went from specific inflexible agents to using files and folders.

I’ve updated my ICM Field Guide, a free reference full of workflow patterns, project examples, and pointers to useful tools and templates. It grew out of Jake Van Clief’s ICM Architect and the ideas shared over Tea and in the Clief Notes community. I wanted those references organized somewhere I could actually use them while working on a project. The two skills work together: - Architect builds and restructures the workspace. - Field Guide is the reference shelf next to it: examples to explore, approaches to compare, and tools to look into. The guide now has 57 Markdown files, including 14 applied patterns and 15 project examples. It also covers workflow kits, template references, decision tools, and implementation tools. The guide includes descriptions and references; the original course materials and template packages stay with their sources. Jake’s introduction to Architect: New tool: ICM Architect Field Guide and installation instructions: ICM Field Guide on GitHub A starting prompt: “Here’s the workflow I’m building. Show me relevant patterns, examples, and tools from the Field Guide.”