Mainstream commercial interest aligning with accessibility needs – double hurrah with extra chocolate on top

Audio Narration

0:00

/441.5738775510204

1×

I dread visiting a new or updated website when I do not have time to ascertain the necessary but often highly opaque combination of tabbing, arrowing and multi-finger key chords for each step in a protracted sequence to download a document, book a table, sign my tax return or verify my identity.

Mrs Mairs assures me that even sighted website useability ranges from ultra clear but rather boring to stunningly beautiful but utterly unusable. Just occasionally a really talented designer makes a site simultaneously a joy to look at and a joy to use. But even that joyousness is frustratingly ephemeral. An urgent business need to announce an amazing new offer or capture more user emails can often break the original design, turning a simple well understood workflow into an unwanted game of Whack A Mole. At other times a website owner commissions the dreaded ‘brand refresh’. Even if the new design is as useable as the old one this re-design imposes a cognitive load on loyal users who thought they knew their way around, much the same as when a supermarket capriciously moves the ‘All Natural’ granola from the cereal aisle to the ‘healthy eating’ aisle.

It is hard to convey how much harder web access is for blind users. Imagine a broken mouse, a screen full of white buttons on a white background interspersed with other buttons labelled ‘Unlabelled Button 1’, and a Help button that does nothing because you haven’t spotted the need to firstly prove you are still not a robot.

Accessibility features making vision impaired access less painful are sadly often not implemented, inadequately implemented or not maintained. And even if accessibility support is legally compliant, the experience can still be poor because the blind user has to pretend they are in some sense sighted.

Serendipitously this is exactly the same predicament that AI agents face when performing actions on the web for a human user. Some early agents   actually leveraged the accessibility features designed for unsighted access but have increasingly found these features to be unreliable or inadequate (need I say more). So most AI agents also pretend to be sighted humans searching out and clicking elusive clumps of pixels, with endless patience, a characteristic of  which I am notoriously bereft.

Because agents have no memory from one visit to the next, they are constantly exploring websites and struggling to click on exactly the right location with varying success. This is slow and cumbersome. So here’s the hurrah; as sighted users increasingly delegate web access to personal AI agents, huge $$$ will be won or lost depending on how well a website enables those millions of personal agents to access the site’s services. Anticipating this changing behaviour, Microsoft and Google recently announced WebMCP, so that agents perceive each co-operating web site as a  set of tools rather than an intrinsically visual ‘human first’ interface.

These personal agents may be deeply integrated into your favourite browser (think Gemini for Chrome or Co-pilot for Edge), or may be provided by a foundation model lab like Open AI/Anthropic, or may be one of the new breed of highly personalised micro claws, functionally similar to the virally adopted OpenClaw, and capable of running on anything from a  tiny Raspberry Pi to a Mac or Windows laptop. 

For vision impaired access this is extremely exciting. Creating a truly blind friendly   claw with an exclusively audio interface is now just a case of software engineering – a task that in 2026 can be completed in a matter of days, if not hours, by your favourite coding agent. This ‘voice first’ claw can, via Chrome or Edge,  predictably access every website that implements WebMCP. 

Fortunately upgrading literally billions of web pages for first class agentic access via WebMCP may be more tractable than it appears at first sight. 

• everything continues to work just like today for pages that are not yet updated – sighted users get their usual experience, agents pretend to be sighted users and blind users swear a lot.
• The majority of pages already follow templates or have a static framework with dynamic content, so only the templates and frameworks need updating, not the ever-changing content itself.
• The WebMCP standard is, at the moment, clean, simple and elegant, meaning many sites can be updated transparently using an automatic script, in theory  without any recourse to design committees, compliance reviews or other sources of friction.

To be clear, wide scale adoption of WebMCP may still not fix accessibility for the vision impaired. Google, Microsoft and Apple must at the very least allow users to connect their own custom agents or claws to the WebMCP support in Chrome, Edge and Safari respectively. If the browser providers refuse to unbundle the browser agent from the browser we are  dependent on their goodwill to make the integrated agent (e.g. Gemini or Co-pilot) fully accessible. This feels like groundhog day, although there are only a few browser vendors and surely at least one of them will either do a great job of accessibility or support unbundled personal agents. So I’m hopeful that: 

• agentic access will be important enough for major site owners and framework vendors to add WebMCP support 
• there will be  enough openness on the browser to connect truly blind friendly conversational voice claws to the WebMCP service. 

Whatever happens, WebMCP is not a quick fix. If the AGI maximalists are right, we will have AGI many years before we have ubiquitous WebMCP and universal accessibility. I cannot speculate what the world looks like if and when we do have genuine AGI. But in the meantime, with help from the brilliant Simon Willison, we are prototyping a Chrome Extension connecting a  conversational voice claw to early WebMCP enabled websites. If that goes well, we will engage a major organisation for whom accessibility really does matter, such as the UK government, to drive meaningful early adoption of this important new standard.