
Your product’s next user may not be a person. In the business I work in, the first agents to show up were not ours. We have built automations, but we haven’t yet put a true agent into production, and I am comfortable being a fast follower. The franchisees in our network are not waiting. They are building their own agents for business management and reporting, and we help them do it. It is only a matter of time before students do the same.
Most product teams are planning how they will use agents. I wrote last week about who has to change for your own agent to work. This is the other side of that question: what happens when other people’s agents arrive at your product, booking, buying, cancelling and comparing on someone’s behalf, without ever seeing the screens your team designed.
Few teams have decided. That decision is being made anyway, by default, in each company’s bot rules and login flows.
Three doors
The shopping agents of the past few weeks show the options, because retailers have already had to pick one:
- Close the door. Amazon began blocking agents from making purchases.
- Open it to approved agents. Walmart launched an Agent Gateway for approved agents, on terms Walmart sets.
- Let the agent decide. The Verge reported that a Meta agent kept re-adding The RealReal’s listings to carts each time a 20-minute hold expired, blocking other shoppers for hours or longer. The company’s chief technology officer then said its bot defences make it much more difficult for all agents to log in, add items to cart, and check out.
The third door is the one most product teams are standing at without knowing it. The RealReal didn’t choose its agent policy. An agent chose it, by breaking the checkout, and the company responded by shutting everyone out.
Keeping agents out keeps customers out
My instinct is for openness, because people choose to use agents. A student who asks an assistant to book a class, or a franchisee whose agent pulls the month’s figures, is a customer doing what customers do, with a new tool. Block the agents and you have blocked the people who sent them.
Cory Doctorow wrote that it’s easy to come up with amazing applications for technology if you’re allowed to assume that you can get everyone else in the world to set aside their priorities and their interests in favor of yours. When agents reach your product, you are the everyone else. Agents only work if the product on the other side cooperates, which gives you more power over the outcome than the agent’s maker has, and a choice about how to use it.
I want agents that help the people in our centres and our students make faster progress: the student learns more, and our people spend their time with students rather than on admin. An agent that books, reschedules and reminds is doing exactly the work I would rather no person had to do. Shutting it out would be protecting the admin.
Charging belongs in the equation
Openness without a price is a subsidy, and a business change has to be sustainable. An agent that checks your booking system again and again, all day, costs you something a person never did. One that bypasses your screens also bypasses everything you built into them: the upsell, the reminder, the moment where a person would have noticed something was wrong.
So the open door needs terms. Some agents get in free, some are rate-limited, some pay, and some are sent to an API instead of the screen, which is cheaper for both sides. Walmart’s gateway is the clearest example so far of a company treating agents as a channel with a contract, rather than as traffic to block or tolerate. It is the same question I asked about learning in the travel agent test: when a new channel sits between you and your customer, who owns the relationship?
Too early to decide?
The strongest objection is timing. Agent traffic is tiny, the standards don’t exist yet, and any rule written now will be wrong in six months. Block the bad bots and wait.
Waiting is also a decision, made by whoever configured the bot blocker. The rules don’t need to be final. They need an owner: someone who decides which agents get in, watches what they do, and changes the terms as traffic grows. Without that person, the first agent to break something sets your policy for you.
Those terms come down to four decisions:
- Which agents: none, approved ones, or anyone that identifies itself.
- What they may do: read, compare, book, buy or cancel.
- On what terms: free, rate-limited, paid, or through an API instead of the screen.
- What stays human: the moments that matter so much to the relationship that you want the person there.
Block the agents and you have blocked the people who sent them. Let them in, on your terms and at a price that lasts.
Leave a Reply