--- title: "Twenty Years of Filing Bug Reports Against the Web, and the Machines Finally Showed Up to Agree" canonical: "https://current.tinyfish.ai/issue/49/echoes/article/40509/twenty-years-of-filing-bug-reports-against-the" publication: "TinyFish" publisher: "TinyFish" language: "en-US" guide: "https://current.tinyfish.ai/llms.txt" issue: 49 section: "echoes" section_name: "Echoes" published: 2026-10-01 word_count: 1403 source_links: 9 --- # Twenty Years of Filing Bug Reports Against the Web, and the Machines Finally Showed Up to Agree *October 1, 2026* [Read on the website]() *The web's accessibility tree is a semantic representation of page content that browsers compute from the DOM. It was built over decades so screen readers could interpret interfaces designed for sighted users. In 2026, it turns out to be the same data structure that AI agents query when they need to understand a web page. The function call `page.accessibility.snapshot()` returns roles, names, and states. Not pixels, not CSS. Almost exactly what a screen reader has consumed for thirty years.*[^1] *We spoke with Roz Labelledby, an accessibility engineer whose career spans the full arc from Microsoft Active Accessibility to the current moment, where agent infrastructure companies keep calling her about "machine-readability." She has been working on this problem since before most of their engineers were born.* *She joined the call from her home office, browser DevTools open on a second monitor showing an accessibility tree. "I always have it open," she said. "It's like weather radar for someone who's been through a lot of storms."* --- ## You've been doing accessibility work for over two decades. When did you first notice that agents were consuming the same infrastructure you'd built for screen readers? **Roz:** Before the agent people did. About three years ago, someone from an automation company asked me to consult on why their browser tool kept failing on certain pages. I pulled up the accessibility tree, showed them the missing labels, the `div`s cosplaying as buttons, and they looked at me like I'd performed a magic trick. I said, "This is what I do. This is what I've *always* done." They had independently arrived at the accessibility tree as their primary interface and had no idea there was an entire profession that had been maintaining and fighting over that tree for decades. They'd reinvented the problem. Just hadn't reinvented the people. ## Take us back to the earliest version of this. Before the web, before ARIA, what did machine-readability even look like? **Roz:** Screen readers in the early days were essentially reverse-engineering the operating system. In Windows 3.1, they'd intercept GDI drawing calls, literally watching what the OS was painting on screen and trying to infer meaning from pixels.[^2] Heroic, fragile work. Every OS update could break everything. When Microsoft introduced MSAA in 1997, it was supposed to fix this by giving screen readers a proper API to query.[^3] But MSAA only worked if application developers *chose to implement it*. And most of them didn't, because, and I'm quoting the sentiment of roughly every product manager I've ever met, "that's a lot of work for a small user base." The pattern was set right there. You can define the API, publish the spec, write the guidelines. None of it matters if the developer doesn't write the label. ## Then the web arrived. HTML has semantic elements built in, so presumably things got better? **Roz:** *[laughs]* The early form spec, RFC 1866, didn't even have a `