I want to talk about something nobody in the AAC world is saying yet, and nobody in the AI world has noticed.
AI is functioning as augmentative and alternative communication for autistic people. Not the kind of AAC you are thinking of — not speech-generating devices or picture boards for nonverbal children. A different kind. The kind that bridges the gap between how an autistic brain organizes information and how the neurotypical world demands it be delivered.
And it is doing it without a prescription, without a therapist, without a stigmatizing device strapped to your chest. It is doing it inside the same tool everyone else uses to write emails and summarize meetings.
That is not a coincidence. That is the best thing that has happened to autistic communication in decades.
What AAC Actually Means
Augmentative and Alternative Communication is a clinical term for any tool or system that helps a person communicate when their natural capacity does not meet the demands of their environment. The field has existed for forty years. Speech-language pathologists specialize in it. Insurance companies sometimes cover it. It has its own conferences.
But the field has always defined "communication difficulty" narrowly. If you can produce speech — if you can physically say the words — you are not considered a candidate for AAC. You can talk. What is the problem?
The problem is that speech is about 15% of communication.
The rest is tone. Timing. Register. Subtext. The ability to calibrate your message to the emotional state of the listener. The instinct to know when to soften a direct statement, when to add social padding, when to laugh at something that is not funny because the power dynamic requires it. These are not communication skills in the clinical sense. But they are the entire operating layer of neurotypical social exchange.
And for autistic people, that operating layer is manual. Every calculation that neurotypical people run in the background, we run in the foreground. Every social adjustment that costs them nothing costs us processing cycles and working memory.
That gap — between having the content and being unable to deliver it in the expected format — is the communication disability that nobody issues a diagnosis code for.
The Translation Tax
I have written about the Admin Tax before — the executive function overhead that eats neurodivergent productivity. But there is a subset of that tax that deserves its own name. I am calling it the Translation Tax.
The Translation Tax is the cognitive cost of converting your native thought structure into neurotypical-compatible output. It is what happens between having the answer and saying the answer in a way that does not get you labeled as rude, cold, blunt, or "lacking emotional intelligence."
Here is what it looks like in practice:
I know the project is going to fail. I can see the structural flaw. The correct communication is: "This architecture will not scale past ten concurrent users and needs to be rebuilt before launch." That is accurate and actionable.
But that is not what neurotypical protocol accepts. Neurotypical protocol requires me to: acknowledge the team's effort first. Soften the critique with qualifiers. Frame the flaw as an "opportunity." Suggest a meeting to "explore options." Express confidence that "we can figure this out together." And then — buried inside four paragraphs of social padding — deliver the same information I could have said in one sentence.
That translation is not free. It costs me thirty minutes and a measurable amount of cognitive battery. For a neurotypical colleague, it costs nothing — they do not even notice they are doing it, the same way you do not notice yourself breathing.
Multiply that by every email, every Slack message, every meeting comment, every client call. That is the Translation Tax. And until two years ago, there was no tool that could pay it for me.
AI Pays the Tax
Here is what I actually do now.
I write what I mean. In my native format — direct, structured, dense with information, stripped of social ceremony. Then I hand it to the AI and say: "Make this appropriate for a client email." Or: "Soften this for a team that is going to be defensive." Or: "Add the social layer I am missing."
The AI does not judge the input. It does not tell me I am being too blunt. It does not require me to mask. It takes my raw signal and wraps it in the protocol the recipient expects.
That is AAC. That is exactly what AAC is supposed to do — bridge the gap between what the communicator can produce and what the environment requires. The fact that it does not look like a medical device does not make it less of one.
And it goes further than email tone.
- Working memory scaffolding. I can dump a complex, nonlinear thought into a conversation and ask the AI to organize it into a sequential argument. My brain does not think sequentially. The world requires sequential output. The AI handles the conversion.
- Executive function bridging. "I need to do X but I cannot start. Break this into steps small enough that I can begin." That is not a productivity hack. That is a clinical intervention for task initiation failure, delivered without a waitlist or a copay.
- Social script generation. "I need to decline this invitation without damaging the relationship. Write me the message." The AI knows the social formula. I do not have to reverse-engineer it from first principles every time.
- Emotional translation. "I wrote this and I think it sounds angry. Does it?" Having a neutral reader that can flag tone mismatches before I send the message has prevented more relationship damage than a year of therapy.
None of these functions appear in any AAC product catalog. None of them are covered by insurance. None of them require a prescription. And every single one of them is doing more for my daily communication capacity than any clinical tool I have ever been offered.
The Invisible Ramp
Here is why this matters beyond my personal workflow.
Traditional AAC carries stigma. A child with an iPad communication board is immediately marked as different. An adult using a speech-generating device in a meeting is immediately recontextualized from "colleague" to "person with a disability." The tool works, but the social cost of using it is enormous. Many people who could benefit from AAC refuse it because the stigma outweighs the utility.
AI has no stigma because everyone is using it. When I ask Claude to rewrite my email, I am doing the same thing my neurotypical coworker is doing — just for a different reason. She is saving time. I am translating between cognitive architectures. The tool is identical. The need is different. And nobody can tell from the outside.
This is the invisible ramp. The building entrance that does not have stairs at all, so nobody knows who needed the ramp and who did not. The accommodation that is indistinguishable from convenience.
Universal design advocates have been arguing for this for decades: when you build the accommodation into the infrastructure instead of bolting it on as a special case, it stops being an accommodation and starts being architecture. AI did this for cognitive accessibility without anyone planning it.
What the AAC Field Is Missing
The clinical AAC community is still thinking about AI as a feature inside their existing products. Better word prediction. Smarter symbol selection. Faster text-to-speech. These are useful increments, but they miss the structural shift.
The structural shift is that mainstream AI tools are already doing AAC work for millions of neurodivergent people who would never walk into a speech-language pathologist's office. People who can talk just fine but cannot navigate the social layer without burning out. People who have the ideas but cannot sequence them into the format their boss expects. People who understand the content but cannot produce the performance.
These people do not have a diagnosis code for their communication gap. Many of them do not have a diagnosis at all. But they found the tool anyway, because it was just sitting there in a browser tab, and it did the thing they have been failing to do manually for their entire lives.
That is the population the AAC field has never served. Not because they did not want to, but because the clinical model could not reach people who technically "can communicate" but do so at a cost that compounds into burnout and career stagnation.
The Cognitive Prosthetic
I want to push this further, because AAC only covers the communication piece. AI is also functioning as a cognitive prosthetic for executive function, and that distinction matters.
An autistic brain does not lack the ability to plan, organize, prioritize, or initiate. It lacks the ability to do these things on demand, reliably, in the format and timeframe the environment requires. The capacity is there. The on-ramp is broken.
AI provides the on-ramp. Not by thinking for you — I am not outsourcing cognition. I am outsourcing the formatting of cognition. The difference is the same as the difference between a person who cannot walk and a person who cannot walk on stairs. The second person does not need a wheelchair. They need a ramp.
When I ask AI to break a project into steps, I am not asking it to think about the project. I already know the project inside and out. I am asking it to convert my three-dimensional mental model into a one-dimensional task list, because the world runs on task lists and my brain runs on spatial maps.
When I ask it to remind me of context from earlier in a conversation, I am not asking it to understand for me. I am compensating for the working memory limitations that come with a brain that processes everything at high resolution and cannot always hold all the threads at once.
These are prosthetic functions. Not intelligence augmentation. Prosthetics.
Why This Is Not Being Talked About
Three reasons.
First, the people benefiting most from this do not have the language for it. They know AI helps them. They cannot articulate why it helps them in a way that distinguishes their use from everyone else's. They say "it saves me time," because that is the socially acceptable explanation. The real explanation — "it translates my brain into a format the neurotypical world can receive" — requires a level of self-disclosure most autistic adults are not prepared to make at work.
Second, the AAC research community and the AI research community do not overlap. AAC researchers are in speech-language pathology departments studying symbol-based communication for nonspeaking populations. AI researchers are in computer science departments studying model capabilities. The Venn diagram has almost no intersection, so nobody is studying the thing that is happening right in front of them.
Third, acknowledging this would require admitting that a consumer product built for profit is doing more for cognitive accessibility than decades of clinical intervention. That is an uncomfortable truth for a healthcare system that has consistently failed autistic adults. It is easier to ignore it than to confront what it implies about the clinical model.
Bilingual, Not Broken
There is a better frame for all of this than "assistive technology," and I want to name it directly.
Autistic people are bilingual. We are fluent in our own cognitive language — the pattern-dense, associative, high-resolution format our brains actually run on. And we have learned, to varying degrees of fluency and at enormous cost, to produce output in neurotypical protocol. We switch between these two languages constantly. The exhaustion is not from the thinking. It is from the code-switching.
What AI does is not fix the way we think. It does not smooth out the autism. It gives us agency over the translation. It lets us stay fluent in our native language — the one where the ideas actually live — and choose when and how to bridge to the other one. That is not a crutch. That is what bilingual people do when they use a dictionary or a translation app: they remain who they are in one language while gaining access to another.
The difference between a tool that changes your thought and a tool that amplifies your control over how your thought travels is the entire difference between assimilation and agency. Masking is assimilation — you suppress the native signal and produce a synthetic one at personal cost. AI-mediated communication is agency — the native signal stays intact, and you choose the format it arrives in on the other end.
This is what the AAC field should be studying. Not whether AI can generate better symbols on a communication board, but whether mainstream AI tools are enabling a new model of neurodivergent communication — one where the person retains full ownership of the thought and outsources only the packaging. That is not a clinical intervention. It is a civil right: the right to think in your own language and still be understood.
What I Am Actually Saying
I am not saying AI replaces therapy or clinical support. I am not saying throw away your diagnosis and get a ChatGPT subscription. I am saying that for a specific and enormous population — autistic and otherwise neurodivergent adults who can speak but cannot communicate in the format the world demands without unsustainable cognitive cost — AI is functioning as the most effective assistive technology that has ever existed.
And it happened by accident. Nobody at Anthropic or OpenAI sat down and said, "Let's build AAC for autistic adults." They built a general-purpose language model, and it turns out that a machine that excels at translating between communication styles is precisely what you need when your brain speaks one language and the world requires another.
The wheelchair was not designed for any specific person. It was designed for a problem: the human body sometimes cannot traverse the built environment. AI was not designed for autistic communication. It was designed for a problem: humans sometimes need help translating between formats. That the translation is between cognitive architectures rather than spoken languages does not make it less valid.
The Protocol: If your brain produces signal in one format and the world only accepts another, the tool that handles the conversion is not a luxury. It is AAC. And for the first time in history, that tool is sitting in everyone's browser, costs twenty dollars, and carries no stigma. Use it without apology.
