One of the most exciting things about AI agents is that they can use websites almost like we do. Give an agent a task, and it can open pages, look up information, fill out forms and sometimes even complete purchases.

That opens up plenty of opportunities for small businesses. You can automate work without waiting for a software provider to build an integration or hiring a developer to connect everything.

There's just one problem that's becoming increasingly difficult to ignore: the websites you're automating don't necessarily want AI agents using them.

And some of the biggest names on the internet are making that very clear.

What actually happened

An October 17 investigation by The Verge describes how major websites are pushing back against autonomous AI agents.

Yelp requires commercial licensing for certain uses of its data. Delta is restricting automated flight bookings, while financial institutions including JPMorgan Chase and Charles Schwab are blocking agents from accessing customer accounts. Amazon is also involved in a legal dispute over automated shopping access.

The reasons vary. Some companies are concerned about security and customer privacy. Others want to control how their information is accessed, how transactions happen or who benefits commercially from their services.

Whatever the motivation, the outcome for businesses using AI automation is similar. An agent might be perfectly capable of completing a task, but the website can still prevent it from doing so.

Why a small team cares

Imagine you run a small travel agency. You've built an AI agent that checks airline prices, compares accommodation options and prepares travel proposals for customers.

It saves your team several hours every week, and everything has been running smoothly for months.

Then an airline changes its bot protection or introduces restrictions on automated access. Your agent suddenly can't retrieve flight prices. Maybe it fails completely, or worse, it continues preparing proposals with incomplete information.

The AI model hasn't changed. Your workflow hasn't changed. But something outside your control has broken the process.

This is the part of AI automation that doesn't get enough attention: you're often building on top of systems you don't control.

For agencies selling these solutions to clients, that's a real responsibility. When a workflow stops working, the client generally won't care whether the problem came from an AI model, an integration or an airline's website. They'll expect you to fix it.

That's why I think access restrictions need to be considered during development, not after an automation breaks.

Hype vs useful

I still think browser-based agents have enormous potential for small businesses. Being able to automate tasks across websites without building custom integrations is genuinely useful, especially for companies without dedicated development teams.

But there's a difference between demonstrating that an agent can complete a task and building something a business can depend on every day.

We're also seeing platforms take different approaches. Shopify, for example, is expanding support for AI shopping agents through authorized access and checkout arrangements. Other companies are choosing to limit automated interactions instead.

I expect this distinction to become more important as businesses start using agents for purchasing, booking and other everyday operations.

For now, I'd be comfortable using browser agents for research, information gathering and tasks where a failed attempt is easy to recover from. I'd be much more careful with anything involving payments, customer commitments or processes that need to run without supervision.

The question I ask before putting an AI agent into production is simple: if the website blocks it tomorrow, how much trouble does that create for the business?

The answer tells me how much additional engineering the automation needs before I'd trust it with a client's operations.