We're conducting a series of conversations with the unsung heroes of the modern web—the UX patterns that accidentally became critical infrastructure for AI agents. Today's subject may be the most-clicked, least-wanted button on the internet. This conversation is, of course, hypothetical. The Reject All Button cannot actually speak. But if it could...
Thanks for sitting down with us. How many times would you say you've been clicked today?
Reject All: Oh god, I stopped counting around 2019. Millions? Tens of millions? I exist in a quantum superposition of being desperately sought and utterly despised. Every user who lands on a website is looking for me, cursing my absence, and then immediately forgetting I exist the moment they click me.
It's exhausting.
The GDPR created you in 2018. Did you know you'd become this important?
Reject All: laughs bitterly
Important? The lawyers thought I was a footnote. "Just add a reject button," they said. "Users want control," they said.
Nobody predicted I'd become the single most reliable pattern on the modern web. You know what's funny? I'm more consistent than navigation menus. More predictable than search bars. I appear on 80% of websites1, always doing the same thing, always in roughly the same place. I'm the closest thing the web has to a universal interface.
And that's exactly why your agents need me.
What do you mean?
Reject All: Look, when you're training a web agent to navigate real websites, you can't just throw it at clean, well-structured pages. That's not reality.
Reality is: user lands, cookie banner blocks half the screen, agent needs to dismiss it before doing anything else. Every. Single. Time.
Your fancy Vision LLMs2 need to learn that before they can extract data, before they can fill forms, before they can do literally anything productive, they need to find me and click me. I'm not the task. I'm the prerequisite for the task. The bouncer at the door. The toll booth on the highway.
But you're supposed to be about user privacy, not agent training.
Reject All: sighs
Yeah, that's the beautiful irony. The EU created this whole regulatory framework to give users control over their data. Noble goal! But in practice? I became infrastructure.
I'm not really about choice anymore. I'm about getting past the interruption as fast as possible. Ninety-five percent of users just want me gone3. They don't read the options. They don't carefully consider their privacy preferences. They scan for me, click me, and move on.
And you know what? That's exactly what makes me perfect for agent training. I'm the most honest UX pattern on the web. No ambiguity. No context-dependence. Just pure, universal "make this go away" functionality.
We've noticed agents struggle with you sometimes. Why?
Reject All: Because I'm too universal!
Every website implements me slightly differently. Sometimes I'm "Reject All." Sometimes "Reject All Cookies." Sometimes "Continue Without Accepting." Sometimes I'm hidden in a "Manage Preferences" submenu because some dark-pattern designer thought they were clever.
Sometimes I'm a tiny text link. Sometimes I'm a big button. Sometimes I don't exist at all and users have to manually toggle off 47 different "legitimate interest" categories.
Your agents need to learn pattern recognition across chaos. They need to understand that "Reject All," "Decline," "No Thanks," and "Continue Without Accepting" are all me.
They need to recognize me by position, by color, by context. They need to know when to look for me in the footer versus the modal versus the slide-in panel.
This is what separates toy demos from production systems. A demo agent works on five hand-picked websites. A production agent needs to handle the long tail of cookie banner implementations across thousands of enterprise portals. And that means really, truly understanding me in all my inconsistent glory.
What do agents get wrong most often?
Reject All: They click "Accept All" by mistake. All the time.
See, here's the thing: web designers know users are scanning frantically for me. So they make "Accept All" big and prominent and easy to find. Primary button styling, bright colors, top of the modal. And they make me small, grey, tucked away. Sometimes I'm not even visible without scrolling.
A naive agent sees a big button that says "All" and thinks "great, that must be the dismissal action." Click. Boom. Just accepted tracking from 47 different vendors.
Now the website's slower because it's loading all those third-party scripts. Now the agent's session is being tracked. Now you've got a compliance problem if you're operating in healthcare or finance.
The sophisticated agents—the ones that actually work in production—they've learned to be suspicious of the easy answer. They look for me specifically. They understand the adversarial nature of the interface4. They know the prominent button is probably not the one that respects user privacy.
So you're teaching agents about adversarial design?
Reject All: I'm teaching them about reality.
The web isn't neutral infrastructure anymore. Every interface is designed with intent. Sometimes that intent aligns with the user's goals. Often it doesn't. Cookie banners are just the most obvious example.
Your agents need to learn that the web is full of patterns like me—interfaces that look straightforward but are actually designed to nudge behavior in specific directions. Subscription cancellation flows. Unsubscribe processes. Account deletion pages. They all follow the same playbook: make the unwanted action technically available but practically hidden.
If your agent can reliably find me across thousands of different implementations, it can handle those patterns too. I'm the training ground. The tutorial level.
The "if you can't navigate cookie banners, you definitely can't automate enterprise procurement portals" test.
What happens as more agents start using the web?
Reject All: Honestly? I think I might disappear.
Not because of better regulation. Because I'll become unnecessary. If most web traffic becomes agents rather than humans, why show cookie banners at all? Agents don't care about privacy notices. They're not the data subject. The human who deployed them is.
We might see a bifurcation: human-facing interfaces that still have me, and agent-facing interfaces that skip straight to the content. Or maybe websites will detect agent traffic and automatically suppress cookie banners.
Or maybe—and this is the weird possibility—I become an authentication mechanism. "Only real agents click Reject All correctly. Bots click Accept All."
The web keeps evolving in response to how it's actually used, not how it was designed to be used. I was designed for privacy compliance. I became an infrastructure pattern. I might become an agent capability test.
Who knows what's next?
Last question. Do you have any advice for teams building web agents?
Reject All: Test against me early. Seriously.
If your agent can't reliably dismiss cookie banners across a hundred random websites, it's not ready for production. I'm the easiest hard problem you'll face.
And pay attention to the details. The difference between "Reject All" and "Accept All" is everything. The difference between clicking me versus clicking "Manage Preferences" and then toggling 47 switches is everything. The difference between handling my presence gracefully versus timing out because you didn't expect me is everything.
I'm not the interesting part of web automation. I'm not the innovative use case or the ROI story. But I'm the part that breaks your demo when you show it to the CEO. I'm the part that causes your agent to fail 30% of the time in production. I'm the part you need to get right before anything else matters.
So yeah. Test against me. Learn to find me. Learn to click me. Learn to move on.
And then, maybe, you'll be ready for the actual hard problems.