The Solo Operator’s AI Stack: What It Costs and What Breaks
If you run a business by yourself, your AI tools stopped being a rounding error some time in the last year. They are now a real monthly line item that moves without warning, and most of the advice written about them is written by people who never had to pay the bill.
This page is the index to the work on this site that starts from an actual invoice instead of a pricing page. Five pieces, in the order a solo operator actually needs them.
The short version: the published price of an AI tool is not what it costs you. Agentic tools bill by consumption, so the same subscription can produce very different invoices two months in a row. The fix is to measure your own usage first, then decide what to cap, route, or run yourself – in that order.
1. Start with the bill you already have
Before changing tools, find out what the current ones consume. AI Coding Agent API Cost Audit: The Real 30-Day Bill walks a four-week audit: inventory and baseline, measure actual consumption, identify where the waste is coming from, then act on it.
It is the longest piece on the site for a reason. Almost every surprise invoice traces back to consumption nobody was tracking, and the pricing page cannot tell you that.
2. Understand why an agentic tool’s bill jumps
Cursor is the clearest example of the shift from flat-rate to consumption pricing. Cursor Usage-Based Billing Explained: Avoid Overages covers how the credit pool and overage system work, why agents burn credits faster than manual use, and how to set a spend limit before the month gets away from you.
Read this one even if you do not use Cursor. The mechanics are becoming the industry default, not a quirk of one product.
3. Know that today’s prices are the promotional ones
AI Subscription Costs Are Artificially Low makes the argument that current subscription pricing sits below what serving these models actually costs, and looks at GitHub Copilot’s move to usage-based billing as the first domino rather than an isolated decision.
Planning a stack around prices that are still being subsidised is how a business gets caught out. Assume the number goes up and design for it.
4. Cut the bill without cutting capability
How to Reduce LLM API Costs: The Self-Hosted Stack is the structural answer: four layers, covering what belongs in an orchestrator, what a router layer does, where GitHub and Obsidian replace paid SaaS storage, and the point at which a VPS is genuinely worth it.
This is where most of the recoverable spend actually is. Subscriptions get cancelled and re-subscribed; a routing decision keeps paying you back every month.
5. Decide whether to run models yourself
How to Self-Host AI Models: Beginner’s Guide answers the question everyone asks first – whether a GPU is required – then covers hardware tiers by machine, the four tools worth knowing for local models, and an honest account of the tradeoffs against cloud.
Self-hosting is not automatically cheaper. It is cheaper for specific, repetitive, high-volume work, and worse for almost everything else. The guide is explicit about which is which.
How to use these in order
- Measure what you spend now, using the 30-day audit.
- Cap the consumption-billed tools so a bad month cannot compound.
- Route the work that does not need your most expensive model.
- Self-host only the narrow, repetitive jobs where it clearly wins.
Doing it in the other direction – buying hardware before measuring – is the common expensive mistake.
For what this looks like on a real bill rather than a spreadsheet, AI Website for Small Business: A Real 2-Day Rebuild follows one company that stopped paying an agency to change a sentence. The recurring line item it removed was larger than the entire tool stack above.
A note on the numbers
Every price in these pieces was accurate when the piece was published, and each one carries its own date. This category reprices often. Treat the figures as a snapshot and the method as the durable part.
More in Tech & AI, or see the content engine that runs this site if you want to see one of these stacks working end to end.
If the workload rather than the bill is the problem, AI for Small Business Operations is the companion index – the same approach applied to what to automate first.