Agentic AI for Accessibility

Embedding AI into accessibility workflows to help champions prioritize what to fix first.

🤖 Agent design

Product Designer, Agentic UX Lead

Timeline
2025 – ongoing
‍
Team
Product Management, Engineering, AI Research

TL;DR - Impact so far
- Translating remediation goals into prioritized results.
- Ranking findings based on business context like impact, traffic, or legal risk.
- Guiding users with structured, confidence-building flows.

Overview

Level Access helps organizations keep websites and apps accessible at scale. Teams often face hundreds or thousands of findings and don’t know where to start. This project uses agentic AI to, translate remediation goals into prioritized results, rank findings by business context (impact, traffic, legal risk), and guide work with structured, confidence-building flows.

This project explores how agentic AI can cut through the noise by:

Translating remediation goals into prioritized results.
Ranking findings based on business context like impact, traffic, or legal risk.
Guiding users with structured, confidence-building flows.

Agent landing screen introducing AI-powered prioritization. Users can rank issues based on impact, traffic, and feasibility instead of sorting manually.

The Prioritization Agent isn’t just a feature, it’s built as a platform engine that works across dashboards, setup, tables, evaluations, and remediation. I authored Level Access’s first Agentic UX Guidelines, now used to embed AI consistently across the product, and I lead the agentic UX strategy.

The Challenge

Research and discovery identified five recurring problems:

Overwhelming volume of findings “teams don’t know where to start” and “after the easy wins, the backlog feels endless.”
Lack of prioritization insights, forcing reliance on manual scoring or external agencies.
Disconnected workflows between findings, remediation planning, and issue trackers.
Inconsistent prioritization criteria across organizations.
Poor integration with issue tracking software, making prioritization invisible downstream.

Proof of Concept (PoC)

Goal: validate feasibility before investing in productization.

In the PoC, the filters component adapts to user input — when filtering with AI, the bar becomes a full text box.

Built a Prioritization Agent

To rank findings against user goals (e.g., VPAT compliance, high-impact fixes).

Minimal UI

Using mostly existing components, and building the AI design system components on the go.

Focused on backend and logic

Query execution, context handling, and ranked outputs.

Use real data

Confirmed findings could be prioritized efficiently with real data and scale.

MVP

We turned the PoC into a guided experience embedded in the product and proved the prioritization agent could operate as an engine across context windows. We also tested whether users trusted and understood AI-driven prioritization in day-to-day workflows.

Embedded Level AI chat within Digital Asset Findings to make prioritization conversational and contextual.
Validated that open prompting alone failed, users needed upfront options, grouping, sorting, and structured guidance to feel confident.

Designed a chat-like flow with inline feedback (👍 👎) to continuously improve outputs.
Added smart filters (status, severity, URL, finding title, etc.) to give users control while still benefiting from AI guidance.

MVP flow — chat experience with smart filters and inline feedback

Here’s the MVP flow in action, the agent ranks findings through chat, applies smart filters, and the table displays the results and additional features.

Roadmap

As the MVP matured, the focus shifted to scaling the prioritization agent as a platform engine. My role moved beyond feature design to shaping the long-term vision and ensuring consistent, future-proof agentic UX foundations.

Roadmap wireframes — from initial touchpoint to remediation

My involvement‍

Defined the roadmap with PMs for how the agent expands across context windows.
Authored the first Agentic UX Guidelines to make AI touchpoints consistent.
Ran cross-functional workshops to align feasibility with vision and keep delivery moving.

Impact

Turned the agent into a platform capability powering multiple workflows.
Established a shared design language for all upcoming AI initiatives.

Contextual slide-outs within the findings table (MVP chat + integrations support)

Beyond the Feature: Agentic Guidelines

As we embedded agentic AI into workflows, it became clear that we needed a systematic approach for consistency across the product. I initiated and authored the first Agentic UX Guidelines at Level Access, now adopted company-wide.

This work shifted my role from feature design to leading the agentic user experience vision across the company. Core principles include:

Invisible, not absent: AI should blend into workflows, not create parallel ones.
Proactive, not reactive: anticipate needs and nudge at the right time.
Keep humans in control: ensure visibility, reversibility, and choice.
Use familiar patterns: AI outputs should reuse existing UI, lowering cognitive load.
Design for system readiness: structure information architecture and workflows so agents can act reliably.

Agentic UX Guidelines — principles to embed AI consistently across the product.

Takeaways

Looking ahead: as AI evolves, design shifts from humans using AI to agents working alongside humans. If agents create future content, websites, documents, media, our role may be to design systems that are readable and actionable by both people and agents.

AI should guide, not overwhelm, blending into existing workflows.
Designers must get technical enough to partner on feasibility and cost tradeoffs.
The problem space and company needs are as critical as user flows when shaping AI.
Embedding AI as an engine, not a bolt-on, ensures scalable, future-proof solutions.
Success meant thinking systemically: aligning with company goals, and shaping information architecture.

Agent's flow, how can we embbed the logic on all context windows.