PathMate - AI Browser Agent for Accessibility
A voice-and-keyboard-driven browser helper that lets people with motor or vision impairments complete complex multi-step web tasks by just describing what they want done.
Adults with motor, vision, or cognitive disabilities who struggle with dense web UIs
- Voice-driven task descriptions ('book me a dental appointment Tuesday afternoon anywhere near zip 11201')
- Screen-reader compatible step-by-step narration with pause/resume on every action
- Verification step: agent reads back form contents and asks for confirmation before submitting
- Saved task library for routines (renew prescription, check benefits, file expense)
LLM browser agents are finally reliable enough to delegate full multi-step tasks, opening independence gains that were impossible with rule-based screen-reader automation.
Large addressable population (~16% globally with significant disability per WHO) and growing assistive tech market, but no evidence of direct demand for a standalone AI browser agent product — most accessibility discourse focuses on making websites accessible rather than delegating to AI.AI Agents & Accessibility: Building Autonomous Interfaces Everyone Can Use ↗Assistive Products Market Report 2024 ↗
Browser Use, Skyvern, Stagehand, and OpenAI Operator target B2B automation, not disabled end-users; 'Screen Reader AI' exists only as an academic paper, not a shipping product, leaving the accessibility-agent niche open — though incumbents could add it cheaply.Browser Agent Frameworks Compared: Browser Use vs Stagehand vs Skyvern ↗Screen Reader AI: A Conversational Web-Accessibility Assistant for Blind and Low-Vision Users ↗
Assistive tech market is $23.1B growing to $33.4B by 2031 (5.1% CAGR), but disabled users have lower average incomes and tolerance for paid SaaS is constrained — viable paths are B2B2C (employers/insurers/payers) or freemium, and per-task LLM costs compress margins.Assistive Technology Market-Global Growth Analysis 2024-2031 ↗Assistive Products Market Report 2024 ↗
Accessibility needs are durable and demographic tailwinds (aging population, EAA/ADA enforcement) are strong, but risk that OpenAI/Google ship native accessibility features in their own browser agents, or that improved web accessibility standards reduce the need for an intermediary tool.Browser Agent Reliability — Benchmarks, Hype Gaps and What Real Task Performance Looks Like ↗The Compound Accuracy Problem: Why Your 95% Accurate Agent Fails 40% of the Time ↗
Multi-step browser agents have median ~70% reliability with re-prompting required, and compound accuracy (95%×10 steps = 60% success) makes failure likely on complex tasks — for disability users, silent failures carry real harm (wrong purchases, missed form fields), so human-in-the-loop is required, raising build complexity.Browser Agent Reliability — Benchmarks, Hype Gaps and What Real Task Performance Looks Like ↗The Compound Accuracy Problem: Why Your 95% Accurate Agent Fails 40% of the Time ↗