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.

Well... I ran the HTML class 😂 And something happened that I did not really think about when I wrote the first two posts. If you read the other posts, the idea was relatively simple. ChatGPT on my phone would become the live teacher. The classroom on my computer would show me the lesson, images, reading, questions, fill in the blanks, etc. And an ICM structure would sit underneath it all so that lessons and progress don't disappear every time a conversation ends. We built it. And I sat down and took the first real lesson. It worked. The teacher knew what I was seeing on my computer. I could read something on the screen, answer the teacher by voice, click through an activity, and continue the conversation. For the first time it actually felt less like "talking to ChatGPT in Thai" and more like I was sitting in a language class with a teacher who happened to also know exactly what was on the screen in front of me. Except... The lesson was too easy. 😂 I went through all 13 activities in about 6 minutes. Every one of the activities the computer could actually check, I got right on the first attempt. At first you could look at that and say: Great! I passed the lesson. But that turned out to be the least interesting thing that happened. Because while the classroom saw that I got everything right, the voice teacher noticed something different. The lesson was working with the Thai structure ถ้า...ก็... Basically an "if this... then this..." relationship. I could use it. I could make new sentences with it. At one point I even made a sentence about going surfing if it didn't rain that wasn't one of the sentences the lesson gave me. But when the teacher asked me to explain what ก็ was actually doing in the sentence, my answer wasn't nearly as clear. And suddenly we had three different pieces of information. The computer said I passed everything. The teacher said I could use something that I couldn't necessarily explain very well. And I was saying the whole lesson was way too easy.

@Ry Mac posted about helping his dad write a book: record him talking about his life, let Claude turn the transcript into chapters. Good post. Good instinct. I used the same method (with ChatGPT two years ago and published a book on a Yoga Style called BikYasa Yoga https://www.amazon.com/dp/B0D4QTCLZH?lv=shuf&channelId=500&plpRedirect=mhFallback). I read it Ry Mac's post twice, and the second read wasn't recognition. It was closer to déjà vu. I didn't set out to build a team of AI workers. I set out to finish a book that had been stuck for longer than I want to admit — one of several, all half-written, all going nowhere on their own. The actual method came from @Jake Van Clief: one worker, one job. Not one assistant asked to do everything passably. A separate one, built for exactly one thing. So the first two were small and specific: a transcriber that turns spoken audio into clean text, nothing else, and a content editor that takes that text and shapes it into something a chapter can be built from. The reason those two existed was a technique from a writing workshop: answer the book's questions out loud instead of staring at a blank page and typing them. Dictate the answer, the transcriber converts it, the editor refines it into prose. Two workers, one job each, built for that loop and nothing more. The book got finished. Then it needed a cover, so I built a designer. Then it needed somewhere to live, so I built a builder. Then it needed people to actually find it, so I built someone for marketing. None of that was planned before the step before it went stuck. Each one showed up because the real work hit a wall without it — never because a diagram said it should exist yet. Five books came out of that, in forty-five days, before any of it had a name worth putting on a page. The post this morning is describing a discovery.

I think I have 30-40 hours left of Fable access. My usage has just ticked over to 56% used. I had enough of a build queue that I decided to upgrade my plan just for this month and the Fable-in-subscription window and honestly, it's already been worth it. How those tokens have been spent: - 5% used to complete a system recarve and hardening pass - 35% used to orchestrate a full substrate port (from NotePlan to Notion) *and* a full branch that is future compatible with that state ready to switch to on Friday assuming this week on Notion goes as planned (oh and a full school comms rewrite to go with my brand new gmail set up exclusively for my system to access). I didn't have to build a thing in Notion - Sonnet dispatches did the lot. - 5% used to build the work order stage of my construction crew - 11% used to build the orchestration, build oversight and dispatch stage of my construction crew (and a massive shout out to @Ari Evergreen - I forked Pushing Dispatch_ and that helped me keep this to 11% of Fable orchestrating, otherwise it would have been more) Every stage of the construction crew has been pushed through a live run of the step before it in the chain to produce the document that guided the build. For the work order stage, that meant a PRD from the first stage and a build spec produced by the second stage. For this orchestration, it meant that I had fully sliced work orders, mapped to my schema, clustered to work packets that made sense. 8 stages; that were able to run in an orchestration-style mode (on Fable, I would not have let Opus go like that to be honest) and I came back to high quality work, measured to my own rules and standards and only a few small tweaks that didn't require substantial rework before committing. Honestly, the fact that something of this scale could be overseen with only 11% usage feels like a real testament to the whole chain here. I mean... this is really something. That means I still have 44% of my Fable tokens to go and only 30-40 hours to spend it in... and now we see how much of it I can use before I lose it... Wish me luck.

So I don’t have much evidence on the quality of the output of Opus 5.5 yet. It does seem like it’s fixed the disaster of yappy and overthinking Opus 5. The one thing that I have noticed is that it’s light years faster than any previous Opus model. Anyone else seeing that too?