I opened my AI bill. It said $847.
For a one-person food blog. The same operation now runs on $47 a month and a GPU that paid itself off in five weeks.
Want the free bonus pack? Join below.
What is inside
- The Intelligence Spectrum — the three-question filter that classifies every AI task into Layer 1 (commodity, free), Layer 2 (skilled, mostly free), or Layer 3 (judgment, pay for the best). For most entrepreneurs, 80% of tasks turn out to be Layer 1. You have been paying surgeon rates to take blood pressure.
- The Routing Decision Framework — the test-cheap-first algorithm that moves tasks off premium APIs without losing quality, plus the routing table you can copy: every task type, the model that runs it, the quality gate, the monthly volume. It saved the author’s operation roughly $800/month in the first 90 days.
- The Overnight Machine — the batch-and-queue pattern that runs your commodity work on local hardware between 10pm and 6am, when electricity is cheap, your GPU is idle, and free-tier API rate limits have reset. Same work, same quality, done before you wake up — about $1 in electricity instead of $80 in tokens.
Who this book is for
Entrepreneurs running AI at real volume — with or without a GPU. The book is built around three pillars (local hardware, free-tier cloud APIs, and open-source tools) and only the first one requires hardware. If you spend under $300/month on AI APIs, the book recommends skipping the hardware entirely; the free-tier playbook alone cuts most bills by 60–80%. The architecture is technical, the frameworks are not — every chapter has a non-technical action that pays back this month.
Frequently asked questions
Do I need a GPU to read this book?
No. Only one of the three pillars requires a GPU, and Chapter 3 maps every pillar separately. If you spend under $300/month on AI APIs, the book recommends you skip the hardware entirely and route everything through the free-tier playbook. The GPU pays back fast if you spend $500+/month, but it is not the price of admission.
Will this work on a Mac?
Yes, with caveats. Apple Silicon runs Qwen3, Llama, and Mistral surprisingly well — slower than a top GPU but more than fast enough for overnight batch work. If you have an M2 Pro or better with 32GB+ unified memory, you already have the hardware. The one gotcha: a closed MacBook will not run jobs overnight, so the overnight machine wants a Mac mini or a Linux box. Both work.
Is this just for technical people?
The architecture is technical. The frameworks are not. The Intelligence Spectrum, the Routing Decision, the Three Layers — these are managerial frameworks for thinking about where money goes. You can apply them tomorrow without writing a line of code, and that alone saves most entrepreneurs $50–$100/month.
What about quality? Does the cheap model produce worse output?
Yes. Sometimes. For some tasks. That is the entire point. Chapter 4 frames this as the 92% vs 97% problem: for threshold tasks (alt text under 125 characters, a meta description with the keyword) the difference is irrelevant; for gradient tasks (launch copy, strategic analysis) you keep paying premium. The book teaches you to separate the two and route accordingly.
How is this different from “just use the cheap model” or “just batch your calls”?
Those are tactics inside the framework. The book is the framework: which tasks, what the quality bar is, when to escalate, how to spread requests across free-tier rate limits, and how to design failure recovery so a 3am timeout does not block the rest of the queue. Plus the routing table, the template-library pattern, and the morning-review ritual that keep it running for years instead of weeks.
What about the new model that comes out next month?
The Judgment Layer (Chapter 6) is where new frontier models go immediately. The Execution Layer (Chapter 7) churns less. The book treats new model releases as cheap experiments against 5% of your volume instead of expensive migrations against 100% — you adopt the newest model where it earns its premium without re-platforming the other 95% of your operation.
Get the free bonus pack
The book references a free bonus pack throughout — the actual files behind it, not theoretical templates. Inside the Zero-Token Enterprise bonus pack:
task-classification-worksheet.csv— the Intelligence Spectrum mapped to a spreadsheet: list every task, run the Three Questions, get a tier assignment and a routing recommendation in the same rowrouting-table-template.csv— a production routing table pre-populated with 18 of the most common solopreneur task types, ready to clone and edit for your stacktoken-tax-calculator.csv— drop in your current monthly API costs and task volumes, get back your token tax estimate, your projected zero-token monthly cost, and your hardware payback periodovernight-schedule-template.md— the 10pm-to-6am batch schedule with the five-batch ordering pattern and the failure-recovery rulesquality-gate-checklist.md— the per-layer quality-control checklist, one page, printablepat-actual-cost-audit.md— the real before-and-after monthly cost breakdown: the $847 month, the migration, the $47/month after
Sign up and the download link lands in your inbox right away.
Also free: From Zero to Sovereign, the series quickstart PDF, is available to download now.
About the author
Pat Tokuyama runs All Day I Eat Like a Shark — an eight-site content-and-commerce operation spanning a Japanese food blog, a specialty tea shop, and a portfolio of niche sites, plus a YouTube channel with 37,000 subscribers. He runs it solo, with a roster of roughly twenty AI agents doing the work of the team he could not afford to hire, for about $260 a month. This book is the cost architecture underneath that operation — the $847 month, the migration, and the $47/month it runs on now.
Book 3 of The Sovereign Entrepreneur series. Each book stands alone; together they are a curriculum for building a business where you own the machinery instead of renting it. Explore the full series at the Sovereign Entrepreneur hub, and grab the free series quickstart, From Zero to Sovereign.