Let's be honest with each other for a moment.
You hate CAPTCHAs.
Not mildly find them annoying. Actually, genuinely, deeply hate them. The ones where you squint at a grid of grainy images trying to determine whether that pixel in the corner counts as part of a bicycle or not. The ones where you click all the traffic lights and it just generates another grid. The ones where you are absolutely certain you are a human being, you have been a human being your entire life, you have significant evidence of this, and yet some rectangle on a webpage is looking at you with profound suspicion and asking you to prove it. Again.
You have failed a CAPTCHA. Everyone has. You have sat there, mildly offended, wondering whether you are in fact a robot and simply have not been informed.
Here is the irony that nobody talks about.
CAPTCHAs were not built to stop robots. They were built to stop bad robots. The assumption baked into every CAPTCHA ever designed is that the thing on the other side of the screen is either a legitimate human being, who should be let in, or a malicious bot, who should not. Those were the only two options the web had to work with for a very long time.
That assumption is now wrong. And the consequences of it being wrong are going to reshape how the open web works in ways most people have not thought through yet.
A Brief, Slightly Sad History of Proving You Are Human
Cast your mind back to the early days of the commercial web. Spam was exploding. Bots were everywhere. Malicious automated scripts were signing up for email accounts in bulk, scraping content without permission, buying concert tickets the millisecond they went on sale, and generally making a nuisance of themselves across the entire internet.
The response was CAPTCHA. Completely Automated Public Turing test to tell Computers and Humans Apart. A test that, in theory, only a human could pass because it required visual pattern recognition that computers of the time could not replicate.
For a while, it worked reasonably well. Then it did not work at all. AI got good enough at solving CAPTCHAs that the original text-based versions became essentially pointless. So they got harder. More distorted. More obscure. The arms race between CAPTCHA designers and the AI systems trying to crack them has been running ever since, and the humans caught in the middle have been the collateral damage the whole time.
reCAPTCHA, the version most people encounter today, moved to image recognition tasks partly because they are harder for machines and partly because Google was using the responses to train its own AI. So every time you identified a fire hydrant or a crosswalk, you were doing unpaid data labelling for one of the world's largest technology companies. Which is either a brilliant piece of design or a mild outrage depending on your perspective.
But here is where we now stand. The AI systems trying to solve CAPTCHAs are now often better at solving them than the humans they are supposedly designed to stop. The test designed to identify humans is now sometimes easier for machines than for people.
The whole thing has become a bit of a farce. And the rise of AI agents is about to make it completely untenable.
The Agent Problem No One Warned You About
When people talk about AI agents and the web, the conversation usually focuses on what agents can do. Research, purchase, book, manage, execute. The capability story is genuinely exciting and we talk about it a lot.
What gets talked about less is the access problem.
Before an AI agent can do anything on a website, it has to get through the front door. And right now, for a very large number of websites, the front door has a bouncer that is specifically designed to keep out anything that looks automated.
Bot management software sits in front of most major e-commerce sites, content platforms, and web services. Its job is to detect non-human traffic and block it. And it is very good at its job. It looks at behavioural signals, mouse movement patterns, typing rhythms, IP reputation, request frequency, device fingerprints, and dozens of other signals to determine whether what is arriving at the door is a person or a machine.
An AI agent browsing a website looks, to these systems, almost indistinguishable from a malicious bot. Because in both cases it is automated software making requests to a web server. The fact that one is trying to scrape prices for a competitor and the other is trying to buy running shoes on behalf of a legitimate customer is, from a purely technical standpoint, invisible to the bot blocker.
The result is that a significant proportion of legitimate AI agents trying to do legitimate things on the legitimate web are currently getting blocked. Not occasionally. Regularly. They hit the front door, the bouncer looks at them, does not recognise what they are, and turns them away with a 403 error that roughly translates to: you are not welcome here.
This is a problem for agents. It is also, if you think about it, a problem for every website that just turned away a customer.
Why the Old Binary Does Not Work Anymore
The CAPTCHA worldview divided the web into two categories. Humans, who are allowed. Bots, which are not.
That binary made sense when bots were almost exclusively malicious. Scrapers, spammers, credential stuffers, inventory hoarders. The bot was the enemy and the human was the customer and the job of security systems was to tell them apart.
But now we have a third category. An agent that is automated software, like a bot, but acting legitimately on behalf of a verified human being, like a customer. It is not a human. It is not a malicious bot. It is something new that the existing infrastructure simply does not have a category for.
And when your security system has no category for something, its default response is to block it. Because blocking things that do not fit is safer than letting them through.
This is not a flaw in the design of these systems. It is a rational response to a world that has changed underneath them. The design was correct for the world that existed when it was built. The world has moved on and the design has not caught up yet.
Verified agent identity is how it catches up.
What a Cryptographic Passport for Agents Actually Looks Like
Here is the idea at the heart of solving this problem and it is actually quite elegant once you see it.
Instead of asking an agent to prove it is human, which it cannot do because it is not, you give it something better. A verified identity that proves it is a legitimate agent, acting on behalf of a verified principal, with a documented authorisation to be doing what it is doing.
Think of it as a cryptographic passport. Not a human passport. An agent passport. A digitally signed credential that carries verified information about who the agent is, who owns it, what platform it runs on, who it is representing, and what it has been authorised to do.
When this agent arrives at the front door of a website, it does not try to pretend to be human. It presents its passport. The bot blocker, which has been updated to understand this new category of visitor, reads the passport, verifies the cryptographic signature, checks that the information checks out, and makes a decision. This is a verified agent representing a legitimate user with a valid payment credential. Let it through.
No blurry traffic lights required.
The information in that passport matters enormously. It is not just "I am an agent, trust me." It is a structured, verifiable bundle. The platform that built this agent is known and accountable. The user it represents has been through KYC verification with a real financial institution. The payment credential attached to this transaction was tokenised and verified by the card network. The mandate this agent is operating under was explicitly authorised by the user with biometric confirmation.
That is a far richer and more trustworthy signal than a human clicking on crosswalks.
What This Means for Merchants
If you run an e-commerce site or any kind of web service, this shift is worth paying attention to carefully.
Right now you are probably blocking a meaningful percentage of legitimate agent traffic without knowing it. Your bot management software is doing its job, which is to stop automated traffic, and it cannot distinguish between the bad kind and the good kind because the infrastructure to make that distinction does not yet exist at scale.
As verified agent identity becomes standard, that changes. You will be able to configure your systems to allow verified agents through while continuing to block unverified ones. The legitimate customer whose agent is shopping on their behalf gets through. The malicious scraper does not.
This is actually better security than what you have now. Not just equivalent. Better. Because a verified agent with a cryptographic identity is, in some ways, more trustworthy than an anonymous human. You know exactly who it is representing. You know the payment credential is genuine. You know the transaction was explicitly authorised. You know the platform it runs on has agreed to a set of rules and is accountable for its agents' behaviour.
The anonymous human clicking through your checkout can still be a fraudster. The verified agent has a paper trail that goes all the way back to a real person with a real bank account and a real consent record.
What This Means for the Open Web
Zoom out for a moment because the implications here go beyond e-commerce.
The web as it currently exists is built around human visitors. Content is designed for human eyes. Paywalls are designed around human subscribers. Advertising is designed around human attention. Every monetisation model on the open web is premised on the idea that the thing consuming the content is a person who can be shown an ad, persuaded to subscribe, or nudged into a purchase.
Agents blow up every single one of those assumptions.
An agent reading a product review does not see the ads. An agent accessing a news article for research does not contribute to page view metrics. An agent navigating a comparison site does not click the sponsored result because it caught its eye. The entire attention economy, the business model that funds the majority of free content on the internet, simply does not function when the consumer is software.
This creates a genuine question about how content gets funded on an open web full of agents. And verified agent identity is part of the answer.
If websites can reliably identify when the visitor is an agent rather than a human, they can charge differently. A small access fee for agent visits, automatically processed, frictionlessly. Not a paywall in the traditional sense. An agent-readable monetisation layer that lets content providers get paid for the value their content provides to agent workflows, even when that agent never generates an ad impression.
Visa's Trusted Agent Protocol has already flagged this direction explicitly. The infrastructure is being built to allow sites to identify verified agents and request payment before granting access. The agent pays automatically, the cost gets passed to whoever is running the agent, and the content provider gets compensated for the value they created.
This is not the death of the open web. It might actually be its survival mechanism in a world where human eyeballs are no longer the primary unit of web traffic.
The CAPTCHA's Last Days
Let me bring this back to where we started.
The CAPTCHA is not going to disappear tomorrow. There are too many websites running too much legacy infrastructure for that to happen quickly. And for sites that primarily serve human visitors, it will remain relevant for some time.
But its days as the primary line of defence for the open web are numbered.
The future is not a harder CAPTCHA. It is not a more sophisticated version of asking you to click on the traffic lights. It is a verified identity layer that tells the web exactly what kind of entity is knocking at the door, with cryptographic proof to back it up, so that the door can make an intelligent decision rather than just treating everything that is not provably human as an enemy.
That future is being built right now. The protocols are being drafted. The standards are being set. The infrastructure is being laid.
And somewhere, a product manager at a CAPTCHA company is having a very uncomfortable year.
Roving Is Building the Identity Layer That Makes This Possible
Verified agent identity starts with giving every agent a financial identity it can prove. That is what Roving does.
Before any agent on the Roving platform touches money, it goes through KYA verification. Who owns it. What it does. What it is allowed to spend. That verified identity travels with the agent. It is what lets a merchant say yes with confidence. It is what lets a bot blocker distinguish a Roving-verified agent from a malicious scraper. It is what makes the whole system work.
If you are building agents and you want them to operate on the open web without hitting walls, the answer starts with giving them a verified financial identity.
Get early access at roving.money and be among the first to give your agents the identity they need to move freely and safely.
If this piece made you think, or if you have a strong opinion about CAPTCHAs, drop a comment below. We want to hear from you.
Stay in the loop:
Follow us on LinkedIn for product updates, industry news and sharp ideas
Join our Discord community and be part of the conversation shaping the future of agentic finance
Share this with your developer friends who have spent more time than they would like to admit clicking on fire hydrants