AI and Accessibility: What Leaders Say About the Future

The Specialist Trap and Culture

Both women built their expertise inside companies with the resources most organizations envy: dedicated accessibility teams, significant budgets, compliance mandates with teeth. What that taught them was how easily those structures can undercut the thing they were built to achieve. “Dedicated accessibility teams are incredibly valuable, but they can also unintentionally reinforce the idea that accessibility belongs to the accessibility experts,” they said. “It doesn’t.”

Compliance sets a floor. WCAG conformance, procurement requirements and legal standards create accountability that would not otherwise exist. What compliance cannot do is create the culture that outlasts a reorganization. As they put it, you can pass every audit and still run an organization where almost nobody understands why any of it matters. “Accessibility teams provide the expertise,” they said. “Culture makes accessibility last.”

Culture, in their telling, is a function of when people learn. Most people enter the workforce having never been taught how to make what they produce usable by disabled people. Certifications exist, but they tend to reach people who already care. “We shouldn’t have to choose to specialize in accessibility before we learn how not to exclude people.”

Grossman-Kahn offered a small illustration of the alternative. Her son was five, watching her set up a new computer, when he asked how a blind person would know what was happening on the screen before VoiceOver had been switched on. At five he had already absorbed a habit of noticing who might be left out, because it was part of the conversation around him.

The Disability Tax

When that habit never develops, someone else absorbs the cost. In practice, that someone is usually a disabled colleague. Gupta and Grossman-Kahn describe the “disability tax” as the uncompensated work disabled employees take on purely by existing in a workplace not built with them in mind. Their sharper point is that it amounts to two jobs rather than one. Fixing alt text is technical work, quick to name and quick to do. Explaining to a mildly defensive colleague why it matters, again, is emotional labor. That half is slower, harder to log, and almost never addressed.

“Impact is what the disabled employee absorbs, not intent,” they said. Their prescription for redistributing it comes in three parts:

• Educate everyone, not only the people already affected by the gap.
• Build a formal disability advisory structure and pay for it, so feedback runs through a real channel.
• Pay disabled users to test your products, treating it as standard practice rather than a goodwill gesture.

The bill for skipping all three arrives as burnout and attrition among those carrying the load.

AI's Role: Progress and Pitfalls

AI now generates a large share of the alt text, captions and summaries that determine whether content is usable at all. Both leaders are clear that this is progress, mostly on volume. A company with millions of products was never going to hand-write descriptions for each one, and the realistic alternative to a decent generated description has usually been no description.

The limits show up wherever a description requires a judgement call. Should alt text guess at someone’s race, gender or disability status? That is an ethical decision rather than a pattern-matching one, and Gupta and Grossman-Kahn argue it belongs to a person. Strip out human oversight and the errors stop being cosmetic. They start misrepresenting people.

The same split runs through speech technology. Automatic speaker identification and live translation close real gaps. Every one of those tools is also only as inclusive as its training data. If a dataset never encountered a stutter or an unfamiliar accent, that absence does not sit there subtly. A live-translation tool that never saw a user with a cough can fail precisely when someone needs it, for an entirely ordinary reason.

Gupta frames the stakes memorably: “With AI, we’re moving faster than ever but now we have the power to go 1,000 miles an hour in the wrong direction.” That exposure extends to code itself. Microsoft CTO Kevin Scott has predicted that within five years 95% of code will be AI-generated. Models trained without accessible patterns as a default will replicate exclusion across millions of products, shipped by teams who never audit for it. The failure is structural rather than a scatter of isolated bugs.

Leadership Actions and Higher Bar

CEOs hold the highest-leverage moves, according to Gupta and Grossman-Kahn. The first is structural: appoint a Chief Accessibility Officer with real authority, resources, and budget, mandated to spread accountability rather than own the problem alone. The second is operational: fund accessibility initiatives, write accessibility into goals, and recognize teams that build access in from the start. The third is personal and rated highest: model inclusive practices in executive communications. Are charts described aloud? Is meaning carried beyond colour? People notice what leaders consistently prioritize and practise themselves.

The field must move beyond asking disabled people to hiring them and letting them lead. Disabled participants should be present in early discovery and general usability testing, not just audits. Avoid treating disabled users as a single group or a rounding error; a product can work for one user while failing another with layered disabilities. Compliance sets a floor but cannot create culture. The higher bar is building products disabled people love to use, not just can use—preserving their authorship, authority, and agency in every room they enter.