I am asking for a friend. 🤣

I tested Fable 5.1 on animated website design and broke down what makes its best results feel polished instead of AI-generated. You’ll see how I use real design references, layered scroll effects, reusable skills, mobile checks, and screenshot-based verification to guide the build. I also compare Fable 5.1 with Fable 5 and share what changed in speed and cost across my tests. SCROLL-CRAFT SKILL

So, things are moving forward on the startup front. Recovery is going well since the surgery and it is allowing me to do more and more each day. I was able to get my paperwork filed for my LLC through Bizee and setup a VOIP and virtual mailbox until I can afford my own office or co-space. I created a new OS today after getting some ideas from the new Grokbot videos and modified the Karpathy Wiki principles to a slimmer version for my new OS for my business. My Efficient Street Office System is now LIVE. It runs through a tiered system: Chief Of Staff / Specialized Agents w/Subagents and a Librarian the stores and indexes all agent activity. Agents can talk with one another through the Chief of Staff and I can expand. I created it using Vellum w/Claude Code API. The agents are using the following models: - Is it research or fact-finding? → Research Agent (Claude Haiku 4.5) - Does it need writing or editing? → Writing Agent (Claude Sonnet 4.5) - Is it coding or technical? → Code Agent (Claude Opus 4.5) - Does it need thinking/planning? → Analysis Agent (Claude Opus 4.5) - Is it about organizing work? → Project Agent (Claude Haiku 4.5) The Chief of Staff is powered by Vellum, the agent architecture uses Claude Code API. As I was building this, I received a DM from one of the developers. I wrote to him last night disappointed on how quickly the Vellum model was chewing through tokens. He then showed me how I can setup Vellum to use other language models, instead of its own. The difference was night and day on usage. Claude API was still there. I think I spent $1.80 to build the whole system. He then sent me another video on how I can use ChatGPT subscription instead of an API for the LM. This should give much more flexibility and it should be as simple as switching the language model inside Vellum. I'll update you on my findings. Anyway, another productive day of learning and developing in the books. 5 weeks and counting... I hope everyone has a great week ahead.

Claude code was running again today. As I instructed it to help me create a CRM for my new startup. It just finished before my tokens ran out for the current session, so I need to wait till 11:30 p.m. tonight to resume once again. Here's the progress so far. Still needs some tweaks and before I save it as an initial template. I plan to add more features to it as my business needs it.

Alright, I am starting a new business, as some of you already know, in the foundation stages of setting everything up. It is going to be based on a subscription model for monthly AI services and the route that I'm thinking about going is actually setting up individual VMs for each client that way I could control the builds and also keep active accounts active and be able to either shut down or temporarily suspend services if payments have not been completed. Is anyone else running this way, or how do you think I should approach handling an MRR type-setup. I was initially going to use GoHighLevel, But decided I would develop my own tools in order to just gain more experience of actually building stuff instead of just pulling from a service that charges monthly to minimize overhead costs.