What happens when your customer sends their AI?

My Wingman helps me research companies before I spend time with them. I have built it using Claude Code and ChatGPT Codex, and it knows the goals I am working towards. That gives it a reason for the research and helps me decide what deserves my attention.

It has made me think about what a company looks like from the other side of that process. If my Wingman is trying to understand your business, can it work out whether you can help me? Can it find enough information to prepare a useful conversation? And what happens when I ask it to take the next step on my behalf?

On 6 October, Meta and Sierra announced that they are developing Personal Agent Protocol, an open standard for how personal AI agents interact with businesses. The intention is to let a customer's agent work through a company's website, connect to its systems or speak with its own agent, with permissions set by the customer and the business. Sierra plans to publish the first specification later this month.

That is still a proposal under development. But the situation it addresses is easy to recognise: I want something done, and I ask my Wingman to help me get it done.

Imagine asking your agent to find a supplier for something your business needs. You explain the situation and what matters to you. The agent then investigates possible suppliers and comes back with options you can consider.

For the supplier, some of the work that used to happen in a first sales conversation may now happen before anyone gets in touch. An agent could look for evidence that the company understands the problem, check the conditions of its offer and identify questions that still need an answer.

I would want my own business to be understandable in that situation. If someone is looking for help with AI or sales professionalization, they should be able to work out what I help with and how to contact me. Their agent should be able to find that information too. If the description is so broad that it could mean almost anything, I am leaving them to guess.

There will also be things an agent cannot establish from a website. Whether I understand the customer's particular situation is something we have to explore together. A description of my work can help us get to that conversation, but it cannot do the conversation for us.

Sriram Krishnan raised a related question in response to the announcement: how will businesses react when they no longer control the whole experience the customer goes through?

I can see why that would be uncomfortable. A company may have put a lot of work into the sequence of pages, questions and conversations through which it wants a customer to move. A customer with an agent may arrive with specific questions and want to get straight to an answer.

If I have asked my Wingman to investigate a supplier, I want the supplier to help it establish whether there is a fit. I still want room to talk to someone when the situation needs explanation or judgment. The preparation should make that conversation easier to get to.

Permissions matter as the agent moves from research into action. Finding information is different from changing an order or making a commitment. I check what happens on my behalf because I carry the responsibility for it. A business also needs to understand what the customer's agent is allowed to do, and when the customer needs to be involved directly.

For me, this is a useful way to look at my own business. Could a customer's Wingman understand the help I offer, find the answers that are available and bring the remaining questions to me? That is something I can work on now, while the technology for agents to interact with businesses continues to develop.