The Disabled Agents

The other day I was debugging a project and asked my AI agent to investigate – which ended up planting this seed of thought in my head for this writing.

The agent started to hit the wall with some fancy JS rendered documentation page and had to run circles to find a readable source on GitHub.

People spent a decade saying “accessibility is a nice-to-have” or “low priority” and built websites only a fully functioning human eyeball with a mouse could use. Now they want an AI agent to go read those same sites for them, and the agent is standing there like:

“Bro. There’s no content. There’s a div. There’s a spinner. There’s a Cloudflare page calling me a bot.”

And the dev who shipped it is like “why can’t my agent read this???” — bruh, neither can a blind guy, and you didn’t care then either…

The agent isn’t broken, the internet is.

The majority skipped the accessibility bill for 15 years, and now the AI they bought to do their reading for them is handing them the invoice. Now we want to automate everything with widely accessible and cheap help from AI but we can’t because it’s broken.

And yes there are MCPs now and computers fully controlled by the “coworker” agents, which solves the problem on some level, but in a very inefficient way if you ask me. And it’s actually very sad that people now care more about serving machines than their fellow disabled humans.

Yes, I know, we can say the same technology actually can make the broken web accessible for everyone, as I can just speak to the machine and it does the trick. But to be honest I don’t really see people talking about this or anyone pushing this angle.

The takeaway is to keep it simple, focus on functioning and semantic solutions, especially for technical things. Nobody cares if that arrow animates in your API docs. Seriously.

And make websites accessible for the love of Buddha.