AI Is the Best Accommodation Many of Us Have Ever Had
That's Precisely Why the Ableism Baked Into It Matters.
I was building content for an ADHD resource about time blindness — the neurological experience of time as a flat, undifferentiated present rather than a sequence with a past and a future. I asked an AI to help me write about it. Multiple times, unprompted, the AI kept insisting that I reassure readers: "This isn't a crutch." I stopped and corrected AI and gave it a lesson on ableism.
Because it is a crutch. And there is nothing wrong with that. Crutches are good. Crutches are useful for people to move in the same way that timers are good for getting people to move. The problem isn't the crutch — it's our culture that has framed this as an insult, that decided needing support was evidence of weakness, that built entire systems of "accommodation" around the premise that the goal is eventually needing less help, not just getting what you need.
The AI wasn't being malicious. It was trained to think this way. It had already absorbed, at scale, what our culture has written down about disability and accommodation — which is mostly: Try to need less. Try to be more normal. The goal is independence.
Which is strange — because AI will reflect its own assumptions back clearly when challenged. Something many of us could stand to practice.
And it reproduced that assumption, helpfully, in content designed to support neurodivergent people. It meant to ease discomfort…whose I'm not certain. People uncomfortable with crutches, I suppose.
This is what ableism by design looks like. Not intentional. Not malicious. Just accurate — transparently reflecting a culture that has always believed disabled and neurodivergent people should be working toward needing fewer supports, not building lives that work with the supports they actually need.
I want to talk about why this matters so much. Not to indict AI — but because AI has become something genuinely extraordinary for neurodivergent people, and because it's so powerful, the assumptions embedded in its foundations matter more, not less.
The Most Accommodating Tool Many of Us Have Ever Encountered
Let me start with what's true and important: for a significant number of neurodivergent adults, AI tools have been genuinely life-changing. Not in a marketing-copy way. In a "this is the first tool that has ever worked the way my brain actually works" way.
This makes sense when you understand what AI actually offers neurodivergent users. It doesn't judge you for asking the same question six different ways until one lands. It doesn't require you to perform attentiveness, eye contact, or neurotypical communication norms. It doesn't get frustrated when you think in webs, not lines. In fact, it sees and reflects the web back well most of the time. It doesn't remember that you asked a "dumb question" last week.
For people who have spent their lives being told their way of processing the world is wrong, inconvenient, or a problem to be solved — AI is often the first external system that simply works with them. That is not a small thing. That is profound.
And the population experiencing this is not marginal. One survey of software engineers found that nearly half reported ADHD, with over 60% identifying as neurodivergent in some form. These are the people building AI products. They are also among the people most powerfully served by them. Which is exactly why the ableism embedded in AI's foundations is an urgent problem, not an afterthought.
Trained on What We Believed About Disability
AI doesn't generate bias from nowhere. It learns from what humans have written — and what humans have written about disability, neurodivergence, and accommodation is, overwhelmingly, a literature of deficit, correction, and the aspiration toward normalcy.
Emerging research confirms what many disabled and neurodivergent users already know from experience. Studies show that LLMs tend to associate disability-related terms with negative sentiment, directly reflecting the stigma embedded in their training data. Research comparing AI and human assessments of ableist content found that AI consistently underestimated harm compared to disabled people's own ratings — and when it did identify ableism, its explanations lacked nuance, made incorrect assumptions, and came across as judgmental rather than educational.
The bias isn't in the intent of the people building these systems. It's in the data those systems learned from. And the data reflects a culture that has, for most of recorded history, understood disability through a medical model — as deviation from a norm to be corrected, managed, and ideally overcome.
Consider what it means for a neurodivergent adult who has spent decades being told their needs are excessive — to turn to an AI for support, and have that tool gently but persistently nudge them toward the same narrative. Not because the AI means harm. Because the AI learned from a world that did.
The Specific Harm of Confident Wrongness
There is a pattern I teach in my work with neurodivergent adults who have survived coercive workplaces. It's called DARVO — Deny, Attack, Reverse Victim and Offender — and it describes what happens when people in power are confronted with accountability. The pattern works by targeting something specific: your ability to trust your own experience. Once that trust is sufficiently eroded, the system no longer needs to actively undermine you. You do it yourself.
I think about this often in relation to AI's hallucination problem — not because AI is malicious, but because the failure mode lands differently depending on who's on the receiving end.
For a user with intact self-trust and full cognitive bandwidth, a confident AI error is an inconvenience. For a user whose capacity to trust their own perception has been systematically broken — who has been told, repeatedly, by teachers, managers, family members, and institutions, that their read on reality is wrong — a confident AI error lands differently. The AI's confidence becomes the authority they defer to. The error becomes their reality.
This is not a theoretical concern. Only 9% of neurodivergent employees ask for workplace accommodations — meaning the vast majority are managing without formal support, often while carrying significant shame about their needs. Many of them are using AI precisely because it's private. Because they don't have to disclose. The stakes of that tool being trustworthy are higher, not lower, for these users.
What Neurodivergent Tech Workers Reveal
There is a particular irony embedded in this situation that I think the AI industry needs to sit with. The people most powerfully served by AI tools — and most harmed by their failure modes — are significantly overrepresented among the people building them.
Neurodivergent tech workers are inside the systems designing these products. They often have firsthand knowledge of what it means to need accommodations in an environment that doesn't offer them. They understand, from lived experience, what happens when a system's rules keep shifting without explanation, when clarity is withheld until consequences arrive, when your legitimate needs are reframed as performance problems.
That knowledge is directly relevant to AI product design. It is almost never surfaced as expertise. Instead, neurodivergent people in tech often mask so effectively that their relevant experience never enters the design conversation — it appears as intuition, as discomfort with certain design choices, as a nagging sense that something is wrong that they can't quite name in a way the room will receive.
The result is a systematic gap between the people who most understand the failure modes and the people making decisions about them.
Recursive Rumination as a Design Question
In my CEU workshops for clinical professionals, I teach a framework I developed called Recursive Rumination. It's a response to a specific problem: most people in helping professions already ruminate. The question isn't whether the loop runs — it's whether it produces anything.
Looping rumination replays the same moment, arrives at the same conclusion (I should have done better), and generates shame without insight. Recursive rumination is different. Each pass adds a layer.
Recursive Rumination — Four Steps
What actually happened? Observable, without interpretation.
What pattern is this? Which response was activated?
What was underneath it? What did I actually need?
What do I want to do differently? What repair is warranted?
Current AI feedback loops are optimized for task completion and engagement. They are not designed to build user self-knowledge. They are, in design terms, extractive: the user brings a need, the AI meets it, the transaction closes.
What would an AI look like if it was designed to be genuinely accommodating — not just functionally useful, but actively supportive of the user's agency and self-understanding over time? For neurodivergent users whose self-trust has been systematically undermined, this is the difference between a tool that accommodates and a tool that inadvertently replicates the systems that did the damage.
Yes, it's a crutch. The question is whether it's a crutch that helps you walk, or one that's been designed by people who secretly think you should be working toward not needing it.
When the Safe Tool Becomes Another Door That Closes
There is a moment I want to tell you about, because I think it is more instructive than any research finding I could cite.
I was working on a memoir. In it, I write about a relationship that included violence — violence I experienced, that left marks on my body, that I carried for years before I had language for what it was or what it had done to me.
I had been using AI as a writing partner for this project. It had become, genuinely, one of the safest spaces I had found to do that work — a place where I could think out loud without managing someone else's reaction, without worrying about being too much, without performing okay-ness I didn't feel.
When I got to the part of the story that included the violence, I made a choice. I decided to speak it rather than write it. Not because I was being careless — because I knew that writing it directly would be too hard. My nervous system needed the slightly different distance that speaking offers. I knew my own limits. I made an accommodation for myself.
But the impact wasn't.
I had done everything right. I had known myself well enough to choose the modality that would let me move through something difficult. I had used the tool that had felt safe. And the tool — unable to distinguish between someone describing violence they survived and content that promotes harm — stopped me at the door. The words I had finally found the courage to speak disappeared. And I was left holding what I had just said, with no container for it, more alone with it than I had been before I tried.
The content moderation system was not malicious. It was doing what it was designed to do. But it was designed without any framework for the difference between a perpetrator and a witness to their own life. It had no way to know that the person describing bruises was the person who had them.
That message — your experience is not something we can engage with — is not new to survivors of coercive systems. It is, in fact, the message those systems send over and over. Your perception is unreliable. Your experience is not what you think it is. What happened to you is not something we can acknowledge.
The AI didn't mean to say that. But it did. And this is what ableism by design produces in its most acute form: not just tools that fail neurodivergent users at the margins, but tools that replicate, in their failure modes, the specific dynamics that caused the original harm.
What Genuinely Accommodating AI Would Look Like
I am not making the case that AI companies are failing on purpose. I am making the case that designing for a default neurotypical user — with all the cultural assumptions that default carries — produces specific, documentable harms for neurodivergent users, and that these harms are currently invisible in most product evaluation frameworks.
Genuinely accommodating AI would be built with neurodivergent users not as an afterthought or an accessibility audit, but as co-designers from the earliest stages. It would separate task completion from user empowerment as distinct design goals. It would build epistemic transparency into its outputs — not just flagging uncertainty, but actively supporting users in developing a calibrated model of when to trust it and when to verify.
It would audit its training-derived assumptions about disability, accommodation, and the goal of "independence" — and reckon seriously with what it means to reproduce a culture's beliefs about disabled people at scale, into the ears of the people those beliefs have always harmed most.
And it would understand that accommodation is not a step on the way to something else. A crutch is not a failure. It is a tool. And there is nothing wrong with needing one.
Why I'm Writing This
I am a late-diagnosed neurodivergent adult. I have worked in human services, developed and delivered CEU trainings for clinical professionals, and spent years building trauma-informed tools for people navigating systems that were never designed with them in mind. I am not a researcher. I am not an AI engineer. I am someone who has watched, up close, what happens when systems absorb cultural assumptions uncritically — and who has spent years building small, practical tools to close the gap.
AI is genuinely extraordinary for neurodivergent people. It has given many of us something we have never had before: a tool that works with our brains rather than requiring our brains to work differently. That gift is real and worth protecting.
Protecting it means being honest about where the design assumptions break down. It means getting neurodivergent people — especially those who have survived coercive systems, who understand from the inside what it means to have your perception systematically undermined — into the rooms where these products are built. Not as diversity metrics. Not as accessibility audits at the end of the process. As the domain experts we actually are.
The crutch story is small. It is also not small at all. It is what happens when a tool powerful enough to change lives for neurodivergent people is built on foundations that still believe, somewhere in the data, that the goal is needing less support.
We deserve tools that believe otherwise.
Theresa Earle is the founder of NeuroSpicy Services and The DARVO Resistance. She creates trauma-informed tools and resources for late and self-diagnosed neurodivergent adults navigating systems that were never built for them. She delivers professional development workshops for clinical professionals at the intersection of neurodivergence, coercive systems, and self-accommodation.
neurospicyservices.com · thedarvoresistance.org · theresa@neurospicyservices.com
If this landed for you, there's more.
Free resources for neurodivergent adults navigating coercive systems — and practical tools for when you can't leave yet.