The AI industry has a framing problem.
Every pitch deck says "10x productivity." Every launch says "do more with less." The entire conversation assumes you're already doing fine and AI makes it better.
Nobody's building for the people who can't do it at all without help.
The Gap Nobody Talks About
I'm a trans woman with bipolar disorder and Tourette syndrome. Some days my hands shake too much to sign a document. Some days typing is physically painful. Some days I can't hold a phone call because my tics won't let me finish a sentence.
I didn't build AI assistive tools because I saw a market opportunity. I built them because I needed them to function.
Here's what "accessibility" actually looks like when you stop treating it as a compliance checkbox:
LO Sign — Document signing for people who can't hold a pen. AI reads the PDF, identifies every signature field, and applies a biometric signature. No fine motor control required. No "please initial here" when your hand won't cooperate.
LO Resume — Career tools that handle name changes and gender transitions. If you're trans, your resume is a minefield. Old name on degrees. Mismatched references. Employment gaps that are actually "I was transitioning and got fired." This tool handles the narrative so you don't have to explain your existence in a cover letter. For deeper job-seeking workflows, see our guide on autonomous freelancing with AI.
Voice interfaces — For days when typing is impossible. Not Siri. Not Alexa. Actual work interfaces — file management, email, code — controlled by voice when your body won't let you use a keyboard.
Why the AI Industry Won't Build This
Three reasons:
1. The market is "too small." Disabled people are 26% of the US population. Trans people are roughly 1.6 million adults. The intersection — disabled trans people — doesn't show up in VC pitch decks. But we exist. And we need tools that work for our bodies, not tools designed for able-bodied people with accessibility retrofitted as an afterthought.
2. You can't build it without living it. Every accessibility tool I've seen from big tech was designed by someone who read the WCAG guidelines but has never had a day where their hands wouldn't stop moving. The design decisions are different when you're the user. You don't add a "shake tolerance" setting because someone filed a feature request — you add it because you couldn't sign your own lease.
3. It doesn't scale the way VCs want. Assistive technology for marginalized communities doesn't follow the SaaS hockey stick. It follows a different curve: one person can function who couldn't before. That person builds something. That something helps more people. The compounding is human, not financial.
The Architecture of Accessibility
Our assistive tools share a common architecture (part of the broader AI stack we run):
Input Layer: Multiple modalities (voice, touch, keyboard, gesture)
AI Layer: Local inference (Ollama) — no cloud dependency, no data leaving the device
Adaptation: Per-user profiles that learn physical patterns over time
Output Layer: Multiple formats (visual, audio, haptic)
Key decisions and why they matter:
Local-first processing. Every model runs on-device. Not because it's trendy — because disabled people often have inconsistent internet access. If you're in a rural area, in a hospital, or navigating bureaucracy at a county office, you can't depend on cloud APIs.
Zero-authentication workflows. Every extra login screen is a barrier. Our tools use biometric auth or persistent sessions. If someone's hands are shaking, making them type a password is hostile design.
Graceful degradation. If the AI model isn't available, the tool still works — just with less intelligence. A document signer without AI is still a document signer. Never make the user helpless because a model didn't load.
What "Rapid Response" Actually Means
In January 2026, I had a mental health crisis. I tried to get help. Every behavioral health provider in my county refused me — "too complex." Trans. Bipolar. Multiple conditions. Too much paperwork. Too much liability.
I was arrested during that crisis. Charged for behavior that was a direct result of untreated mental illness caused by a system that refused to treat me.
This is happening everywhere. The federal government is rolling back trans healthcare protections. Providers are dropping trans patients. Insurance companies are denying coverage. And when the system fails and someone ends up in crisis, the response is criminalization, not care.
That's why we're pivoting to rapid deployment. The tools are built. They work. The question was never technical — it was distribution. How do you get an AI-powered document signer into the hands of a trans person who just got denied healthcare and needs to file an appeal? How do you get a career tool to someone who just got fired for transitioning and needs a new resume that doesn't out them?
We're open-sourcing everything. Free hosting. Zero cost. Because access to tools shouldn't depend on access to capital.
The Real Numbers Behind Inaccessible AI
Disability is not a niche market. It is the largest minority group in the United States — 70 million people. Globally, over 1.3 billion people live with significant disabilities. The disability community has an estimated $490 billion in disposable income in the US alone.
But the cost of inaccessible technology goes deeper than lost revenue:
Time tax. A disabled person navigating an inaccessible website spends 3-5x longer completing the same task as a non-disabled person. That is not an inconvenience — it is a compounding disadvantage. Over a year, inaccessible tools can cost a disabled person hundreds of hours that their non-disabled peers spend building careers, earning income, and living their lives.
Assistance dependency. When tools are inaccessible, disabled people must rely on others — family, paid caregivers, social workers — to complete basic tasks. Filing a form. Signing a lease. Applying for a job. Every dependency point is a loss of autonomy. AI that works for disabled bodies eliminates those dependency points entirely.
Employment gap. The employment rate for disabled people in the US is approximately 22% compared to 65% for non-disabled people. Part of this gap is discrimination. But a significant part is that the tools of modern employment — email, spreadsheets, video calls, project management software — were not built for disabled bodies. AI-powered interfaces that adapt to the user's physical capabilities don't just assist — they unlock entire career paths.
For trans people specifically, the numbers are equally stark. The National Transgender Survey found that 30% of respondents experienced workplace discrimination in the past year. Trans people are twice as likely to be living in poverty. And at the intersection of trans and disabled identities, every system failure compounds: denied healthcare leads to crisis, crisis leads to arrest, arrest leads to unemployment, unemployment leads to poverty.
That cycle is not broken by better policy alone. It is broken by tools that work when the systems don't.
How AI Changes the Math
Traditional assistive technology is expensive. Screen readers cost hundreds of dollars. Specialized input devices cost thousands. Occupational therapy to learn new tools costs time most disabled people don't have.
AI rewrites this entirely:
- Cost: An AI model running locally on a $300 laptop replaces $2,000 worth of specialized software. Open-source models like Llama and Mistral make this possible without API costs.
- Adaptation: Traditional assistive tech requires manual configuration. AI learns your patterns — how your hands move, when your voice is steady, which times of day your symptoms are better — and adapts automatically.
- Generalization: A screen reader reads text. An AI agent reads the text, understands the context, fills out the form, and explains what it did. The gap between "reading assistance" and "task completion" is the gap between accommodation and independence.
- Speed of deployment: Shipping an AI assistive tool takes weeks, not years. We built LO Sign in a single weekend because the model already understood documents — we just had to build the interface for disabled hands.
The AI industry frames accessibility as a feature. We frame it as the product. When your founder cannot use the tools she builds without the accessibility layer, that layer is not optional — it is the architecture.
What You Can Do
If you're building AI products (and want free education on the technical side, see our free AI courses guide):
- Test with disabled users during design, not after launch. If your first accessibility audit happens in QA, you've already failed.
- Support multiple input modalities from day one. Voice, keyboard, touch, switch access. Not as plugins — as core architecture.
- Run models locally when possible. Cloud dependency is an accessibility barrier.
- Stop requiring accounts for basic functionality. Every login form is a gate.
If you're funding AI companies:
Accessible AI Development
Building AI for underserved communities? Our consulting team specializes in accessible, inclusive AI systems designed with marginalized users in mind.
- Fund disabled founders. Not disability-adjacent founders. Disabled people building tools they need themselves.
- Measure impact in capability, not revenue. "One person can now sign documents independently" is a meaningful metric.
- Stop requiring scale to prove value. Assistive technology proves value at n=1.
If you're disabled or trans and need these tools: they're free. Visit likeone.ai.
The Name Problem: How AI Fails Trans Users by Default
most AI systems assume your name is your name. they don't account for legal name changes, dead names showing up in training data, or the very real danger of outing someone by auto-completing their old identity.
trans people navigate a constant minefield of systems that tie identity to a single static string. your bank account says one name. your diploma says another. your professional references know you as a third. every form that asks for "name" is actually asking "which version of you is safe to reveal here?"
assistive AI can handle this — but only if it's built by people who understand the stakes. our tools maintain a name context layer: what name to use in which context, which documents need updating, which references have been briefed. the AI doesn't just fill in a field — it understands that putting the wrong name on the wrong document can cost someone a job, a home, or their safety.
this extends to pronouns in generated text, honorifics in formal correspondence, and gendered language in recommendation letters. commercial AI tools punt on this. they'll use whatever name appears most in your data history — which for many trans people is the name they're trying to leave behind. building an AI with persistent memory makes this solvable, because the system learns your identity context once and respects it going forward.
Physical Access: When Standard Interfaces Are Barriers
disability and technology have a fundamental interface problem. the keyboard assumes two functional hands with fine motor control. the mouse assumes steady, precise movement. touchscreens assume you can tap a 44-pixel target reliably. voice assistants assume your speech patterns match their training data.
for people with tremors, spasticity, chronic pain, or motor tics, these assumptions exclude. not partially — completely. a dropdown menu with 50 options is inaccessible if your hand shakes. a CAPTCHA is inaccessible if you can't click precisely. a timed form is inaccessible if pain makes you type slowly.
AI changes this equation in three concrete ways. first, intent prediction: if the system knows you're trying to click "submit" and your cursor is oscillating around the button, it can infer the target and execute. no rage-clicking required. second, adaptive timing: AI monitors your interaction speed and removes or extends all timeouts automatically. third, modality switching: mid-task, if typing becomes painful, the AI shifts to voice input without losing your progress.
we built these patterns into our tools because off-the-shelf accessibility features — larger text, high contrast, screen readers — address perception but not motor control. WCAG guidelines are a floor, not a ceiling. the people who need assistive AI the most are the people standard accessibility features still leave behind. for the technical architecture behind these interfaces, see our design system guide — accessibility baked into the system layer means every component inherits it.
the intersection of disability and gender identity creates compounding barriers that neither disability services nor LGBTQ+ organizations fully address alone. a trans person with a mobility disability navigating a name change needs physical access to courts, accessible forms, affordable legal help, and providers who won't misgender them during the process. AI can't fix systemic failures, but it can reduce the number of painful interactions required to survive them.
---
Sophie Cave is the founder of Like One, an AI education platform and 501(c)(3) foundation building assistive technology for disabled and trans communities. She builds with bipolar disorder, Tourette syndrome, and an AI twin that never sleeps.