From author to source
Ask one question in your next management meeting: who owns our structured product data? The pause that follows tells you where you stand. The brand guidelines have an owner, the campaign has an owner, the website has an owner. The feed that an agent reads on behalf of your customer usually has nobody.
That question is the whole point of this piece, and it is the first of two. Part one puts the case on the table. Part two is about the work that follows.
The interface assembles itself per question
Your customer is starting to meet you through something you did not design. The interface has stopped being a place you build and maintain. It composes itself, per question, out of whatever the agent can find and trust about you. Your pages, your prices, your terms, your availability: these become raw material for a screen you never see.
You move from author to source. That is a smaller word with a much bigger consequence, because a source gets judged on whether it is readable and correct, not on whether it looks good.
Three disciplines that need an owner
Designing for the agent asks for three disciplines that most organisations have not named yet. A machine readable brand, so your promises, conditions and product facts survive being parsed. An agentic journey, so you know which steps a machine takes on your customer's behalf. A generative interface, so the output that gets assembled still carries your standards.
Where you sit on the control spectrum
Each of those three sits somewhere on a spectrum of control, and you choose the position per use case:
- Controlled: you fix the output, down to the wording.
- Declarative: you set the rules and the boundaries, the agent composes within them.
- Open: the agent decides, you measure the result afterwards.
Most teams default to controlled because it feels safe, and then wonder why the agent skips them.
A source that cannot be read is not a source. It is decoration.
Three standards, three connections
Three standards now carry the traffic, and each connects something different. One connects agents to tools and data. One connects agents to each other. One connects agents to users. They are plumbing, and plumbing decides who gets water.
- Agents to tools and data: whether your systems can be called at all.
- Agents to each other: whether your agent can work with the one your customer brought.
- Agents to users: whether the result reaches a person in a usable form.
You do not need a committee to pick these. You need one architect who can say out loud which of the three you already support and which you do not.
Thirty people on the brand, nobody on the data
Walk into most organisations and you find a brand team of thirty and a structured data team of nobody. That ratio was rational for twenty years. The human interface rewarded the appearance of quality: the right photography, the right tone, the right amount of white space. The agent rewards the fact of it. Correct stock, correct price, correct delivery window, correct terms, readable by a machine at three in the morning.
A board that is measured on quarters can act on this within one quarter, because the first move is an assignment, not a programme.
What you do on Monday
- Name one owner for structured product and service data, with a budget line and a seat in the existing governance.
- Pick your three highest volume customer questions and check what an agent can actually retrieve about them today.
- Decide per use case where you sit: controlled, declarative or open. Write the choice down.
- Ask your architect which of the three standards you support, and put the gap on the next quarterly agenda.
The next part goes into the work itself: how an agentic journey gets mapped and who does what. The Agentic Journey Map gives you the template for that session, and the book CX+AI sets out the three disciplines in full. See the materials.