We like AI. We’ve integrated it throughout our Accessibility Tracker platform and it does a lot of cool stuff like automatically generate VPATs based on your audit progress.
However, what we continually see in the digital accessibility industry is marketing messaging that deceives consumers about AI’s capabilities. Sometimes this is because accessibility companies themselves don’t understand AI and other times it’s because they’re trying to mislead consumers into thinking their AI product does more than it does.
The accessibility industry has a new favorite marketing angle: AI agents. Vendors are racing to announce “AI-powered agents” that will transform how organizations manage accessibility. It sounds impressive. It’s also not what’s actually being delivered.
What’s being sold as agentic AI is, in most cases, a collection of features that have existed for years with a fresh coat of terminology. Let’s go over why the messaging is wrong.
What “AI Agent” Means
An AI agent is a system that takes autonomous action toward a goal. It doesn’t wait for you to click a button or type a query. It perceives its environment, makes decisions, executes tasks, and adapts based on results—with minimal human intervention at each step.
Think of the difference between a calculator and a self-driving car. A calculator performs operations when you input numbers. A self-driving car continuously perceives road conditions, makes navigation decisions, adjusts speed, avoids obstacles, and routes around traffic—all without you telling it what to do at every moment.
Agentic AI operates closer to the self-driving car. It has goals, takes initiative, and chains together multiple actions to achieve outcomes.
What’s Actually Being Sold
Now look at what accessibility vendors are calling “AI agents”:
AI that summarizes scan results. You run a scan, and the AI writes a paragraph describing what changed since your last scan. This is a prompt. The AI responds when asked, summarizes data it’s given, and stops. There’s no autonomous action, no decision-making, no initiative.
AI that generates code suggestions. The platform flags an issue and offers a code snippet to fix it. A developer reviews it, decides whether to use it, and commits it manually. This is autocomplete—the same thing IDEs have done for years, now powered by an LLM.
AI that answers questions from documentation. A chatbot trained on WCAG guidelines and product documentation responds to your questions. This is a chatbot. Chatbots are not agents. They respond when prompted and do nothing otherwise.
AI that generates reports from your data. You configure parameters, hit generate, and get a formatted report. This is templated report generation—a feature that predates AI entirely, now with better natural language in the output.
None of these tools take autonomous action. None of them make decisions without human initiation. None of them chain together multiple steps toward a goal. They’re AI features, not AI agents.
Deception
Calling a chatbot an “agent” isn’t just imprecise — it obscures what’s actually happening in your accessibility program.
When a vendor says their AI agents will help you manage accessibility, it sounds like something is being done for you. In reality, every one of these features requires a human to initiate an action, interpret the output, and decide what to do next. The AI assists with discrete tasks. It doesn’t manage anything.
This distinction matters because organizations buying these platforms may believe they’re getting more automation than they actually are. They’re still responsible for every decision, every prioritization, and every fix. The AI makes some of those steps faster, but it doesn’t independently take action.
What Agentic AI in Accessibility Would Actually Look Like
Real agentic AI in accessibility would operate something like this:
You point the agent at your codebase or website. It autonomously crawls your pages, identifies accessibility issues through multiple evaluation methods, prioritizes them based on severity and user impact, generates fixes, tests those fixes against WCAG criteria, and opens pull requests (without you taking action).
When a developer merges new code, the agent automatically evaluates it for accessibility regressions, creates tickets for any new issues, assigns them based on code ownership, and follows up if they’re not addressed within a defined timeframe.
The agent monitors your production site continuously. When it detects a new barrier, it doesn’t just log it — it investigates the cause, determines whether it’s a content issue or a code issue, routes it to the appropriate team, and suggests a remediation approach with context about why the issue matters.
That’s agentic. The system has goals, takes initiative, makes decisions, and executes multi-step workflows autonomously.
What’s currently being sold is a human doing all of that work, with AI summarizing results and answering questions along the way.
The Truth
There’s nothing wrong with AI-assisted accessibility features. Summarizing scan results saves time. Code suggestions can accelerate remediation. Chatbots can answer common questions faster than searching documentation.
These are legitimate productivity improvements. They just aren’t agents.
The accessibility industry would benefit from more precision in how these capabilities are described. When everything gets labeled “agentic AI,” the term loses meaning—and buyers lose the ability to understand what they’re actually purchasing.
If a platform’s AI requires you to initiate every action, interpret every output, and make every decision, you’re not buying an agent. You’re buying a tool. Tools are valuable. But let’s call them what they are.
Accessibility Tracker AI
While we don’t have agentic AI yet, we’re working on it at Accessible.org Labs.
In the meantime, we have some very nice AI features inside of Tracker including:
- AI remediation guidance (AI helps your team with fixes)
- AI generated VPATs (using your real audit report)
- AI Project Insights (AI provides insights based on your project progress)
- AI generated progress reports (An executive summary and progress report anytime)
You can sign up for a paid plan and access every feature at AccessibilityTracker.com.
