This week (November 10-14, 2025) I am participating in the W3C International Technical Plenary in Kobe , Japan. And everywhere here, people are talking about “AI Agents . “ And this has a huge impact on digital accessibility.
I’ve been meaning to write this article for ages, ever since I read Léonie Watson ‘s wonderful text about AI agents and accessibility. They hold incredible promise and will completely change the way we navigate. That’s why digital accessibility is so important.
For those who don’t know, AI agents are artificial intelligence resources that perform complex tasks during our browsing. Forget about creating images and translating ChatGPT and other LLMs. What I’m talking about are complex interactions between the agent and the interface of a web application.
AI agents are spreading across the web, primarily in web browsers. There are some proposals for websites to have their own AI agents, but the expectation is that operating systems and browsers will be the main focus.
ChatGPT Atlas and Perplexity are great examples. They are browsers that have a prompt allowing you to ask artificial intelligence to perform actions. In a simple test that Diogo Cortiz showed me, he managed to send a direct message on my LinkedIn simply by asking the AI to do so. He also managed to register for an event using only the browser prompt. It seems almost supernatural, since you can follow what’s happening on the screen. It’s something promising and should be present in our daily lives in the coming years.
But what about accessibility?
Some believe that the magic she performed for Diogo will happen with accessibility, that she’ll take a page full of problems and fix it. It’s not quite like that.
These tools have enormous potential for addressing accessibility issues in problematic interfaces. They can adjust headers, indicate required fields, and even provide detailed descriptions of images. But at what cost?
The servers that process ChatGPT requests, for example, require a lot of water to maintain an adequate temperature and avoid overheating. It is estimated that a request to create a 100-word email uses the equivalent of 500ml of water . This is a high cost that is being incurred. Now, imagine what will happen when AI agents dominate the market?
Performing tasks, such as shopping, requesting documents, or even saying hi to a friend on social media, requires processing power. And this processing costs money (and water). If AI agents also have to fix code with accessibility issues, the time and cost of performing this task will be greater. And it may prioritize a more robust and accessible application.
Another important factor is that perhaps not all agents will focus on accessibility adjustments. Some will be specific to this or will have components for this adjustment. And depending on the distribution, it may be more expensive to use an agent that has to adjust the accessibility of a page or application.
There are already concepts for developing “AI-friendly” applications that allow content to be accessed more easily by the user agents of these tools. This aligns with digital accessibility, since the content needs to be simple and easily accessible so that a person with a disability using an AI agent can request a task that will be executed much more easily.
I remember the lectures of my dear friend and mentor in digital accessibility, Horácio Soares , who in his lectures back in 2010 already talked about how accessibility benefited the “blind billionaire ,” making a reference to the ease with which Google indexes accessible pages. The analogy is the same for LLMs and AI agents.
Furthermore, AI agents promise to be excellent tools for verifying accessibility compliance. Developers and managers can use these resources to test their applications and verify if they comply with accessibility best practices, WCAG , and ABNT standards 17225 and 17060 , as well as make code adjustments. This does not eliminate the need for accessibility professionals. On the contrary, these professionals will be important to ensure that accessibility is truly addressed. In times of artificial intelligence hallucinations, human review is fundamental.
Much still needs to happen before this scenario becomes a reality. Issues of privacy, security, and even ethics are still being debated worldwide. However, given the speed at which things are happening, it shouldn’t be long before we see major changes in our access to the web.
Conclusion
AI agents are arriving and will completely change the way we navigate the web. Using these features directly in the browser will be a huge step forward for accessibility. However, this doesn’t mean that web applications should no longer follow accessibility standards. Their use ensures that agents can perform tasks faster and more effectively, benefiting all users. An application without accessibility will also be a problem for AI agents.
It’s important to remember that when we talk about digital accessibility, we’re talking about access for people. Even if we make our applications user-friendly for AI agents, they will still be tools to deliver results to a person. If we disregard accessibility that benefits people, we may be losing a bit of our humanity.