An Open Letter to Jakob Nielsen: Re: Hello AI Agents: Goodbye UI Design, RIP Accessibility

Mr. Nielsen,

I recently read your article Hello AI Agents: Goodbye UI Design, RIP Accessibility published on LinkedIn and Substack. Previously, you wrote an article claiming that Generative AI would be the solution for both UI/UX and Accessibility. While I appreciate your enthusiasm for both generative AI and AI agents—two technologies that I embrace in all of my work inside and outside of accessibility—I must question why this article was written in the first place. It appears just before Accessibility Conference Season (Axe-con, CSUN, DisabilityIN), and seems designed to stir the ire of Accessibility professionals by providing commentary without genuine consideration of the challenges introduced by these technologies.

Ok, Jakob. Challenge accepted.

The web, in its construction and implementation, remains defined by the people.? It is they who set the expectations for its use, and neither AI agents nor generative AI will replace that constant. What does that mean for agents? Read on, my friend. ??

Your summary:?

“Autonomous agents will transform user experience by automating interactions, making traditional UI design obsolete, as users stop visiting websites in favor of solely interacting through their agent. Focus on designing for agents, not humans. Accessibility will disappear as a concern for web design, as disabled users will only use an agent that transforms content and features to their specific needs.”

My response:?

Nah, bruh.? Nah.

To a significant degree, automation has transformed the web, but it is dangerous to assume that it will render traditional UI design obsolete. Even with automation, we experience technology through our human lens. While technology can shape and define some interactions, human behavior ultimately determines whether those interactions succeed in meeting our goals.

Your comment:

“Focus on designing for agents, not humans.”

My response:

Dude, how would that work? Is it an agent that sits on the other end of a computer screen anxiously awaiting an email from the insurance company about its daughter’s health care? Or is it an agent—a crazed fan—desperate to buy tickets to a concert by its favorite artist (#Tswizzle4life—okay, not really, but now you can’t unsee it)? No. At the other end of a computer is a person at an interface working to attain a goal—searching for a dress, a class, or a product or service. In our current systems, we have modules that facilitate these activities. Even if agents replace most of the tasks involved, the agent still serves the person awaiting the result. An agent’s actions are defined by the person with the need. As such, the product or service MUST be designed with, and operated in consideration of, humanity in order to be a viable product. Human-centered UX and human-centered design must continue to reflect users’ choices, needs, and actions while leveraging technology to more effectively facilitate the fulfillment of those needs.

Assumptions and Counterpoints: Accessibility and AI Agents

The promise of AI agents tailoring experiences to individual needs is compelling, but the assumptions underlying this vision require careful scrutiny. A future where AI agents dramatically improve accessibility is possible—but they will likely need to work in concert with ongoing improvements in native accessibility rather than replacing them outright. Below is a summary of some assumptions and counterpoints derived from your article:

• Assumption: Traditional accessibility methods have failed to deliver an acceptable user experience for disabled users.
• Counterpoint: Substantial progress has been made (though I agree there is still more to be done). Continuous improvements and industry efforts—such as training programs and better development practices—have enhanced accessibility on numerous platforms.
• Assumption: Disabled users will be early adopters of AI agents due to their current challenges with existing interfaces.
• Counterpoint: Not all disabled users may be ready or able to adapt to such systems immediately; additional support will be needed during the transition.
• Assumption: AI agents will be sufficiently advanced by 2027–2030 to reliably transform web content for disabled users.
• Counterpoint: Predicting technological maturity is inherently uncertain. Achieving reliable, safe, and universally applicable AI agents involves overcoming significant technical, regulatory, and ethical hurdles.
• Assumption: Once AI agents are adopted, websites can shift focus from built-in accessibility to serving these agents exclusively.
• Counterpoint: Abandoning built-in accessibility risks creating a two-tiered system that depends on the AI’s flawless performance. While personalization through AI holds immense promise, it may also result in fragmented or inconsistent experiences.

What Actually Works? Leveraging Agentic Frameworks for Accessibility

Agentic frameworks can support and improve upon existing and evolving processes for evaluating and mitigating accessibility issues. These frameworks offer a dynamic, data-driven approach that can, in part, complement traditional accessibility practices. By harnessing advanced algorithms and real-time analysis, agent frameworks can assist in the identification, reporting, and remediation of accessibility issues—and often speed up the process of reporting and resolution. The following points illustrate how agentic frameworks can play a pivotal role in making content more accessible:

• Automated Scanning: Quickly locate common accessibility issues identifiable through code analysis (e.g., missing alt text, insufficient color contrast, and improper semantic structure).
• Consistent Reporting: Generate standardized reports that detail identified issues and provide prioritized remediation steps.
• Real-Time Feedback: Offer immediate insights during development, enabling on-the-fly adjustments to enhance accessibility (this is where accessibility overlays can serve as useful diagnostic tools).
• Continuous Monitoring: Track changes and newly introduced content to detect emerging accessibility problems over time.
• User Interaction Simulation: Emulate diverse user interactions to test the accessibility of interfaces for various disabilities.
• Automated Remediation Suggestions: Provide code snippets or configuration adjustments to help developers quickly address specific issues.
• Data-Driven Insights: Aggregate and analyze accessibility data to inform strategic improvements and prioritize high-impact changes.

Conclusion

Mr. Nielsen,? I appreciate the opportunity to engage in this critical debate. Your perspective on the future of accessibility and AI has sparked essential discussion about the limitations arising from primary reliance on both Generative AI and Agentic implementation of technologies to address deeply human-centered challenges.

While I am a strong advocate for innovation, I firmly believe that true progress lies in a balanced approach—one that harmonizes traditional, human-centered accessibility practices with the powerful capabilities of emerging AI tools.

P.S. If you are using an agent to scan this content, I hope it accurately reflects the inaccuracies of the “generative AI/agents will solve everything” position. ??If there are actual inaccuracies in this article, Agents did it :)

Respectfully,

Dr. Keith