Something big is happening in the space between businesses and their customers: an AI now stands there.
It quotes your prices. It answers your complaints. It makes commitments on your behalf. It refuses you a refund at 2 a.m. without asking anyone.
And here is the rare part. The habits of this new relationship have not hardened yet. For a short while longer, both sides still get a say in how this is supposed to work.
That window is genuinely unusual, and I do not think most people notice they are standing in it. Think about how the rules of retail were settled: by the time anyone wrote consumer protection law, the practices were already a century old and the law mostly ratified what shops were doing anyway. The same happened with call centres, with online checkout, with app store refunds. The habits form first, quietly, in millions of small interactions. The rules arrive later and mostly describe what already exists.
We are in the "habits form" phase of AI services right now. What companies do this year — whether they let customers reach a human, whether they leave any record of what was said, whether they admit what the bot is — will become the industry default. Not because anyone decides it, but because that is what everyone will have gotten used to.
That is why I wrote not one guide, but two. One conversation, seen from both ends of the counter.
For those who build and run AI services: Your Company's AI Through the Customer's Eyes. How to bring in AI without quietly losing the people it was meant to serve — what to promise, how to keep it, and how to show you kept it. It is written in the voice of a demanding customer, because that voice is missing from every AI deployment meeting I have ever seen.
For those who meet these bots as customers: Demand the AI Receipt. What a company's AI owes you, which questions to ask it, and what to do when the answers do not satisfy you. Every chapter ends with a message you can send to a company almost as-is.
Both are free, no email, no registration. That is deliberate: a rulebook that only one side can afford is not a rulebook, it is an advantage.
Writing them as a pair taught me something I did not expect. The two documents kept converging on the same three artifacts — public AI rules, a service passport, a receipt — approached from opposite directions. The business version asks "what should we publish so customers trust us?" The consumer version asks "what should we demand so we are not helpless?" Same three answers. Every time.
That symmetry is the strongest argument I have that these are the right three. When the party with power and the party without it independently arrive at the same list, you are probably not looking at somebody's preference. You are looking at the actual shape of the problem.
There is a version of the next few years where AI service quality becomes a real discipline — with standards, independent checks, and customers who know what to ask for. And there is a version where it becomes another layer of the same old asymmetry, just faster and cheaper to operate. The difference will not be decided by regulators. It will be decided by whether enough people on both sides of the counter bother to name what good looks like, while the naming is still possible.
Both guides are in the Library. Take whichever side of the counter you are standing on. Ideally, read the other one too — it is the same conversation.