I’ve spent more than thirty years building identity and fraud prevention systems. During that time we’ve gone through multiple technology shifts—mobile, cloud, biometrics, passkeys, machine learning. None of them fundamentally changed who was using online applications.
Until now.
The security industry keeps describing AI agents as “better bots.” I think that’s the wrong mental model.
AI agents are not bots on steroids. They’re humans on steroids.
Bots automated clicks. Agents automate goals.
A bot follows a script. An agent reasons. It explores, retries, changes strategy when something fails, compares alternatives, and keeps working until it achieves the objective it was given. In many cases it behaves more like a determined human than like the automation we’ve spent twenty years blocking.
That distinction matters because it breaks the assumptions behind almost every customer-facing security product.
For years, life was simple. Humans were good. Bots were suspicious. Fraud prevention was largely about separating the two. Now there is a third population.
Agents acting on behalf of legitimate customers are already researching products, comparing prices, opening accounts, applying for loans, filing insurance claims and completing purchases. Increasingly, your best customer may never visit your website. Their agent will.
If you treat every automated interaction like a bot, you won’t just block attacks. You’ll block revenue. The real challenge isn’t identifying automation. That’s becoming the easy part. The difficult question is understanding whose objective the agent represents, what authority it has, and whether its behavior should be trusted.
The same technology that helps a customer shop for a mortgage can also help a fraudster test thousands of synthetic identities. The difference isn’t that one is an agent and the other is a bot.
They’re both agents. The difference is identity, authorization, intent and risk. This is why I believe fraud prevention has fundamentally changed.
Historically we optimized one metric: stopping fraud. Today every security decision has two outcomes. You either stop fraud without affecting the customer. Or you create friction that quietly destroys revenue.
Every unnecessary challenge, false decline or abandoned login is effectively a tax on growth. Most organizations measure fraud losses with incredible precision. Very few measure how much revenue their security stack prevents them from earning.
Agents amplify this problem because they operate at machine speed while pursuing human objectives.
A customer’s shopping agent won’t patiently retry tomorrow after being blocked. It will simply recommend a competitor. Your security dashboard may celebrate another blocked automated session while your business loses a customer it never realized it had.
That’s why visibility is becoming more important than blocking.
Organizations need to understand how much of their traffic is already agentic, which platforms those agents originate from, who they represent, what permissions they hold, and whether they’re helping a legitimate customer or pursuing abuse.
Without that visibility, every policy becomes guesswork. The winners in the next generation of digital commerce won’t be the organizations that block the most automation.
They’ll be the ones that can distinguish between beneficial and harmful agents in real time, allowing one while stopping the other. For the first time in decades, customer identity is no longer enough. You also need to understand the identity of the intelligence acting on the customer’s behalf.
That isn’t bot management. It’s an entirely new security problem.
