Then it advanced toward me and the eyes changed to a “NO LOITERING” screen.

A few minutes later, inside the store, I turned a corner in the lumber section and a loud voice said “STOP. You are not authorized. This is a restricted area.” A screen I hadn’t even noticed lit up over a door, showing live video of me standing there.

I’m a pretty tech-enthusiastic person. But in about five minutes I got two previews of the future, and neither one felt good. There’s something specifically unsettling about a non-human thing that’s designed to look friendly turning aggressive on you. Add that to the surveillance we’re all quietly living under now — cameras in the Fred Meyer parking lot, Flock cameras on the drive over — and it starts to feel less like convenience and more like something else.

As I sat with it, I realized the thing that bothered me wasn’t the AI part. It was the authority part. The robot wasn’t helping anyone. It was deciding whether I was welcome, and it defaulted to treating me like a threat. That’s a very different job than “answer a question” or “summarize a document.”

Most of the AI conversation right now is about capability — what these tools can now do. The more useful question, especially if you’re the person responsible for a brand or a website, is what you’re comfortable letting them do to the people who show up. Capability and authority are not the same thing, and the industry keeps blurring that line on purpose. Every recent product update from the major AI companies is chasing “agentic” AI — systems that don’t just answer, they act, decide, and move on without checking in. That’s a great pitch in a keynote. It’s a much bigger question when it’s sitting on your website talking to your customers.

Here’s where that shows up in ordinary client work, and what’s worth deciding on purpose instead of by default.

Chatbots that gatekeep instead of help

A lot of website chat tools are sold on deflection — keep people out of your inbox, resolve things automatically, reduce the load on your team. Useful, in the right dose. But it’s worth checking what happens at the edges: does your chatbot ever tell someone no? Does it quietly end a conversation, decide a question isn’t worth escalating, or make a real person feel like they’ve hit a wall with no way through to a human? That’s the same shape as the robot at Lowe’s — a friendly-looking interface making a judgment call it was never really equipped to make.

Takeaway: Use AI to answer and route. Don’t let it be the last word on whether someone gets help.

Forms and fraud filters that treat visitors as suspects

Bot protection, spam filtering, lead-scoring — all reasonable, all increasingly AI-driven, and all capable of quietly deciding a real donor, patient, or customer isn’t worth their time. If your forms are getting more aggressive about who they let through, it’s worth asking who’s actually getting filtered out, and how you’d know. A form is often someone’s only way in the door. Worth checking who it’s turning away by default.

Takeaway: Anything that screens out real people deserves a human periodically checking its judgment, not just its uptime.

Personalization that decides who sees what

AI-driven personalization can tailor content, offers, or pricing to who it thinks someone is. That’s a step beyond helping someone find what they need — it’s making a judgment about them and acting on it without asking. For a business, especially a nonprofit or a mission-driven brand, that’s worth a hard look. Are you personalizing to serve people better, or are you letting an algorithm quietly sort them?

Takeaway: Personalize the experience. Don’t let the system make decisions about people it never explains to them.

Agentic features vendors are pushing right now

Nearly every platform update this year is some version of “let the AI just handle it” — auto-respond to inquiries, auto-qualify leads, auto-escalate or auto-close. Some of that is genuinely useful. But agentic, by definition, means less checking in with a human before something happens. That trade-off is worth making on purpose, feature by feature, not accepting by default because it shipped in the latest release.

Takeaway: Before you turn on an “autonomous” feature, name out loud what decision it’s now making without you.

At XO, we use AI plenty — mostly for research and recurring tasks. It’s genuinely great at that, and it makes the work better. What it doesn’t do, and what we’re not interested in handing it, is any moment where the job is deciding whether someone belongs, matters, or gets a yes. That’s still ours.

That’s really the whole idea. Using AI to help you think or move faster is one thing. Letting it decide who’s welcome is another. Somewhere in your website, your intake process, or your customer-facing tools, that same line exists. It’s worth finding it and drawing it on purpose — before a vendor’s next “agentic” update draws it for you.