Your customers have hired AI agents. Will they find a wall or a front door?

Personal AI agents have moved past answering questions. The wave started with GrokBot. Agents like Instinct, Muse and Dots now act for the people who use them. They log in to accounts, fill in forms, chase providers and make purchases. A lot of that happens without a prompt.

Robbie Tilleard, Lorikeet's General Manager for EMEA, walked through what this means for customer experience teams.

What makes a personal agent different

Someone using ChatGPT or Claude asks a question and gets an answer. A personal agent takes a different approach:

• It's proactive. It wakes up each day, checks the goals its user set and takes action. Nobody has to ask it.

• It has its own computer. It opens a browser in the cloud, goes to your site and clicks through the same steps a person would.

• It can act as the customer. Depending on what the user hands over, it can log in to their accounts, use their card and make calls.

Instinct read one user's email, saw he had a flight coming up and checked him in. He never asked it to. Muse cut another user's cable and phone bills by $800 a year by contacting the provider and following up until the job was done. As Robbie put it, agents are relentless in pursuit of a goal.

“Personal agent traffic already accounts for 5% to 10% of volume on some sites.” - Robbie Tilleard, Lorikeet General Manager for EMEA

Wall or front door

Every business now has a choice to make about these agents. One restaurant booking app banned a user's account after his agent checked for tables every hour. Amazon has stopped AI agents from shopping on its site.

Banning agents today is like banning credit cards online in the early days of the internet. Card payments felt risky then. They became one of the biggest growth engines online. The task is to keep the experience safe for customers while letting in what is likely to become a major channel within the next twelve months. If you block the agent, you lose the customer it was acting for.

Three things to do instead

1. See them

Start with data, even if you're cautious about agents. Separate agent contacts from bot noise. Tag the contacts an agent wrote or drove, log which agents arrive and find where they fail. This is the one step every business should take now.

Lorikeet sorts AI traffic into four types:

• Automated: a message clearly sent by a script or system

• Self-declared AI: an agent that says what it is, such as "I'm Robbie's Instinct"

• AI-drafted: a person used AI to help write the message

• AI-written: the message appears to be fully written by AI

2. Build for them

Make your processes agent-friendly. If a person has to get through five pages of forms, give an agent a faster route to the same result. Over time, agents will favor the businesses that are faster and easier to deal with.

3. Talk to them

An agent browsing your site won't know your rules unless you tell it. Most personal agents are built to follow the rules, but they need them stated plainly. For instance, if customers can't send more than $10,000 a day and you don't say so clearly, an agent might try to send $8,000 one hour and $3,000 the next. A direct, authoritative answer stops that. A help article the agent may never read won't.

Talking to agents also helps when a question is too complex for AI to handle. Your concierge can reply to the agent and offer a callback or an email follow-up. The customer gets a proper answer. Otherwise the agent moves on to a competitor.

Setting it up

Robbie set up Lorikeet B2A for a sample money transfer business in four steps:

1. Publish a concierge

2. Enable the agent endpoint

3. Turn on the agent protocol for your domain

4. Paste a snippet into your site, the same way you'd add analytics

Lorikeet's Coach walks you through each step. You can also set it up from a coding tool like Claude Code or Codex once you're connected. From there, you can see which agents arrive, what they ask and where they get stuck. Each agent gets an answer you control, so it doesn't have to work things out by crawling your site.

Where to start

Your customers have already hired agents, and you decide whether those agents meet a wall or a front door. Start with the data, run small experiments and build from there. Teams that start now will be among the first to do this well, and they'll have real data to bring to internal discussions about agents.

Lorikeet customers can turn on B2A through their forward deployed team. Everyone else can start a 30-day free trial and see what agents are already doing on their site.