Notes for owners · July 2026

Before you buy any AI tool, ask these seven questions.

The AI market is optimized to demo well. Every tool looks miraculous for four minutes on a sales call. The way to cut through it isn’t technical knowledge. It’s the same buyer’s discipline you’d bring to any major hire or vendor. Seven questions, asked plainly:

• 1. Who owns it? If you stop paying, what still works? Rented capability is a hostage arrangement, so get the answer in writing.
• 2. Where does my data live? Your cloud or theirs? Who else can see it, and what is it used to train?
• 3. What stops it from making things up? “It’s very accurate” is not an answer. Ask what rule prevents invented numbers, and how it’s enforced.
• 4. Will it write in my voice, or a house style? Ask to see output trained on a real business’s documents, not a demo prompt.
• 5. Who writes the rules, and who approves them? If the answer is “the model handles it,” there are no rules.
• 6. How is it tested before launch? Against what? Your real work, or a benchmark you’ll never see?
• 7. What does leaving look like? The exit terms tell you more about a vendor than the sales deck does.

What the squirm tells you

Good vendors answer these quickly, because the answers are the product. Evasive ones reveal the business model: lock-in, data leverage, and capability that impresses in demos and guesses in production. You don’t need to know how transformers work to be unfoolable. You need to insist on the same things you’d insist on from a hire: standards, testing, and accountability.

Our answers, for the record

You own it outright, in your cloud, from day one. Your data stays yours. A written rule bars invention; voice is trained on your documents; you approve every rule; every build is tested against your real work and launches when it earns your signature; and if you leave, the system keeps working. It stops expanding, not working. The rest is in the FAQ, and the specifics for your business get mapped on the consultation.