TL;DR
- An AI tool is a tenant, not your house: it belongs to its vendor, and what decides how much it costs you when it changes is what you control.
- Three controls within your reach: the account in your name, your prompts saved outside the app, and a tested second model for the task you'd hurt most to lose.
- Three questions for any vendor or agency: which model and version they use, whether the account is in your name, and whether they tell you when they change the model.
- This week's one thing: pick the AI tool you'd hurt most to lose and answer three questions about it.
The short answer. You can't control what your vendor does. You can control what you're left with when it does. An account in your name, your prompts in a document you own, and a second model you've already tested are what separate an annoying Tuesday from a Tuesday when you don't know how to work.
If you own a boat, you know how this goes. You pay for the slip. The dock belongs to the marina, and the marina sets the hours, the rate, and which rules change in January. You don't own the dock; you own the boat. Now ask the uncomfortable question: do you know which engine your boat runs and who holds the keys?
The same goes for the AI tools you pay for, and almost nobody thinks about it until something changes.
Why a tool is a tenant and not your house
An AI tool is another company's product. That company decides what to add, what to take out, and what to charge. It's normal for any software, and nothing wrong with it. What weighs is when your business starts leaning on it without anyone having written down what happens if it changes.
This very week there's an example in plain sight. On November 5, Anthropic announced that Claude is adding a personal finance section, in beta and only in the US (we covered it in Monday's roundup). It's not a criticism: it shows that the product you use today isn't the one you had a month ago, and sometimes it changes in ways you didn't ask for.
And on November 2, the same company published FlashProve, a research prototype for proving with math which model produced an answer. Today it only covers contexts up to 1,024 tokens, so it isn't something your vendor can offer you. What it does leave behind is the underlying question: which model is behind what you pay for?
You won't settle that with math. You settle it with three controls.
The three controls: account, prompts, second model
This is Hub365 AI's method.
Control 1. The account is in your name
Look at whose email opened the account and who pays the subscription. If an employee opened it with a personal email, or an agency with theirs, someone else holds the key. If that person leaves, your account leaves with them.
What to check: the owner's email, who pays, and whose name is on the invoice.
Control 2. Your prompts and workflows live in a document you own
If what you ask a tool, and how you ask it, only exists inside the app, the day it changes or you leave it you'll lose hours of work. Copy your good prompts into a document of your own, with one line on what each is for. It isn't glamorous and it works.
What to check: can you open a document right now and see the three prompts you use most, without opening the tool?
Control 3. You have a tested second model
Pick the task you'd hurt most to be unable to do tomorrow. Try it in another assistant, with the same input text, and save how it went. You don't need to pay for anything new to start. You need an afternoon.
A tested second model doesn't mean a second contract. It means that if the first one changes, you already know which would be next and what you'd have to adjust.
Where the line sits: these controls aren't about which permissions you give a tool, but about what you keep when it changes. For the first, see two assistants can get into your apps.
The three questions for any vendor or agency
If a third party runs an AI tool for you, ask these three. Not as an accusation: as the normal conversation that should exist.
You don't need to know about models to understand the answer. It's the marina question: which engine the boat runs, who holds the key, and whether they tell you when they swap it.
How to test a second model in an afternoon
If the text includes customer data, check where your client data lives before pasting it into another assistant.
What this plan doesn't solve
- It doesn't tell you which tool is best. It compares none, and names none as the one you should use.
- It doesn't stop a price increase. It gives you something to react with, not a discount.
- A second model costs time, and sometimes a second subscription. Do it only for the task you'd hurt most to lose.
- It doesn't predict what will happen with regulation or with any vendor. None of this claims anything will change. It's about being ready if it does.
The dock belongs to the marina. The boat is yours. Your account, your prompts, and your second model are what you take with you when the rules change.
Keep reading
- This week in AI: what happened November 2 to 6 - Monday's roundup
- Two assistants can get into your apps. You choose which ones - what permissions you give an assistant
- Where your client data lives - before pasting data into another assistant
- The map of the five SEO disciplines - how customers find you today
- Introducing Personal Finance in Claude, Anthropic - the full announcement
Sources
- Introducing Personal Finance in Claude, Anthropic, November 5, 2026 - https://claude.com/resources/articles/introducing-personal-finance-in-claude
- FlashProve (provable inference), Anthropic, November 2, 2026 - https://www.anthropic.com/research/provable-inference
