As the web begins to evolve into something more conversational and action-driven through AI usage, questions about accessibility are once again coming to the forefront. For people with disabilities, especially those who rely on assistive technologies, the promise of agentic AI, the kind of artificial intelligence capable of taking meaningful action on a user’s behalf, offers both new opportunities and familiar concerns.
Take online shopping, for example. For blind users, navigating a department store is only realistically possible with the help of a personal shopping assistant, something most stores don’t offer. And so, as a result, many turn to online retail, hoping for more autonomy. Unfortunately, even the most accessible retail websites often fail to provide the information needed to make confident purchase decisions.
The problem isn’t always the mechanics of browsing such as searching, adding items to a basket, or checking out, but rather the lack of meaningful product descriptions. Key visual details about fit, texture, pattern, or cut are frequently locked away in images, with minimal or no accompanying alt text.
In the absence of adequate descriptions, many blind users now turn to generative AI tools like ChatGPT or Claude to describe images. Some screen readers now offer direct access to these tools, allowing users to request image descriptions in context, but of course, there is still a long way to go to accommodate disabled users, as Léonie Watson (pictured), Co-Founder at TetraLogical explores...
Building Accessibility into Ecommerce
The growing scale and complexity of personalisation in online fashion retail has reached a point where it can no longer be effectively managed without the support of AI and automated systems. However, there are valid concerns about AI; its environmental impact, the ethics of data sourcing, and of course, the well-known risk of hallucinations. But for many disabled users, inaccurate AI descriptions are often no worse than having no information at all. The use of AI becomes a pragmatic choice, a calculated judgment call in the face of limited alternatives.
Still, relying on AI in this way is far from seamless. It depends on how a website is structured, whether images can be detected, and if they contain any metadata or cues that help AI generate accurate descriptions. It’s a time-consuming process that rarely matches the convenience of simply seeing the product.
This is where the concept of the agentic web comes in. Imagine an AI personal shopping assistant that can be asked to find clothing based on specific parameters, provide consistent product descriptions, answer follow-up questions, and even add items to a shopping basket, all through voice or text commands. This is not hypothetical; it’s already happening.
Platforms like Innosearch are pioneering this approach. The service allows users to shop across hundreds of online retailers, presenting products in a more consistent and accessible format. More notably, it includes an agentic AI called CoBrowse, which can take real actions on a user’s behalf: filtering search results, excluding stores, selecting sizes, applying coupon codes, and more. While it stops short of completing purchases for security reasons, it replicates many aspects of a real-world shopping assistant.
Intentional Agents: Toward a More Mindful AI Future
While agentic commerce is still emerging, this layered model reflects my perspective on how it can scale successfully. Powering agentic commerce at scale requires multiple interconnected layers working in harmony, each critical to enabling intelligent, adaptive and autonomous customer journeys.
This shift prompts a broader question: if AI can handle tasks on a user’s behalf, why rely on traditional websites at all? For users who struggle with visual overload, constant motion, or complex navigation structures, an agentic interface may offer a far superior experience.
Beyond individual convenience, there’s evidence of a wider trend. AI use is surging,OpenAI reported 2.5 billion ChatGPT prompts per day as of July 2025, up from 1 billion in December 2024. While Google’s AI search is reducing traffic to websites by as much as 50% in some sectors. And agentic platforms, even on a smaller scale, are part of this shift.
The implications for accessibility are significant. If websites become data sources for agentic systems rather than user-facing platforms, what happens to existing accessibility standards? How can developers ensure that dynamically generated content meets legal and ethical requirements, especially when AI responses vary from one prompt to the next?
And Finally…
Agentic AI isn’t just a convenience, it’s an equalising technology. For disabled users, it offers a chance to engage in digital commerce on their own terms, with far less friction and frustration. In the long term, agentic interfaces may become a preferred method of interaction not just for accessibility, but for anyone who values efficiency, personalisation, and simplicity.
Retailers who invest in AI-ready product data, accessible APIs, and inclusive design practices will be best positioned to serve this expanding mode of interaction, and meet both ethical and legal responsibilities for digital inclusion.
While the agentic web holds enormous potential, especially for people with disabilities, it also presents new challenges. Ensuring that this next iteration of the web is inclusive will require careful design, testing, and accountability. The technology may be evolving, but the responsibility to make it accessible remains unchanged.
