Here is something nobody warns you about when you start using AI as a thinking partner for long enough: eventually it reads all the data, and it tells you something you were not ready to hear.
I have been writing this log for months. Dozens of entries about autism, about masking, about sensory processing, about executive function, about the nervous system architecture that runs underneath everything I do. I have been feeding all of it — the patterns, the burnout cycles, the support structures, the accommodations, the daily cost of functioning — into conversations with AI as I process and write. And at some point, after absorbing enough of it, the model looked at the aggregate and said, essentially: John, based on everything you have described, your support needs are consistent with Level 2.
And I sat there and did the only reasonable thing.
I put my face in my hands. Then I panicked a little. Then I thought about it for a long time.
The DSM-5 Severity Levels
For anyone unfamiliar, the DSM-5 categorizes autism into three severity levels based on support needs:
Level 1: "Requiring support." Difficulty initiating social interactions. May appear to have decreased interest in social interactions. Inflexibility of behavior causes significant interference with functioning in one or more contexts. Difficulty switching between activities. Problems of organization and planning hamper independence.
Level 2: "Requiring substantial support." Marked deficits in verbal and nonverbal social communication skills. Social impairments apparent even with supports in place. Limited initiation of social interactions. Reduced or abnormal responses to social overtures from others. Inflexibility of behavior, difficulty coping with change, or other restricted/repetitive behaviors appear frequently enough to be obvious to the casual observer and interfere with functioning in a variety of contexts. Distress and/or difficulty changing focus or action.
Level 3: "Requiring very substantial support." Severe deficits in verbal and nonverbal social communication skills cause severe impairments in functioning. Very limited initiation of social interactions. Minimal response to social overtures from others.
I was diagnosed as Level 1. The clinical shorthand for the version of autism that looks, from the outside, like it is basically fine. High-functioning. Independent. A little quirky, maybe, but managing.
And for thirteen years, I accepted that framing without examining it too closely.
The Gap Between Appearance and Load
Here is the problem with the severity levels as they are typically assessed: they measure what the clinician sees in the room. They measure presentation. They do not measure the weight of the infrastructure required to produce that presentation.
My life does not look dysfunctional. I built a career. I run a business. I am married. I live on 35 acres. I write and publish. I ship code. By every external metric, I am a Level 1 success story — someone whose autism is real but manageable, whose support needs are modest, who functions independently with minor accommodations.
But here is what that picture leaves out: the support systems are not minor.
My wife functions as my social API layer. She handles the phone calls I cannot make. She manages the social logistics I cannot track. She reads the room when I cannot read it. She translates the unspoken social information that I miss entirely — not occasionally, but constantly, in every context that involves other humans. Without that layer, my social functioning does not degrade slightly. It degrades substantially.
My environment is engineered. Thirty-five acres is not a lifestyle preference. It is a sensory accommodation. The distance from neighbors, the control over sound and light, the ability to manage my sensory environment completely — that is load-bearing infrastructure. Remove it and the sensory overload is not a mild inconvenience. It is disabling.
My routines are structural. The morning protocols, the work patterns, the rigid scheduling, the inability to tolerate disruption without significant distress — these are not personality quirks. They are the scaffolding that holds the day together. When someone or something breaks the routine, I do not get mildly annoyed. I lose executive function. The whole system stutters.
My AI tools have become assistive technology. Not in the casual sense of "they make things easier." In the clinical sense of "they compensate for deficits that would otherwise prevent independent functioning in these domains." Communication processing. Social script generation. Executive function support. Task initiation. Without them, the output drops — not by ten percent, but by magnitudes.
Deficits That Persist Even with Support
This is the phrase from the Level 2 criteria that stopped me cold: social impairments apparent even with supports in place.
Because that is accurate. Even with my wife translating. Even with my engineered environment. Even with AI assisting my communication. Even with every accommodation and support system I have built over decades — the deficits are still there. They are managed. They are compensated. They are worked around. But they persist.
I still cannot initiate social interactions without significant effort and frequently avoid them entirely. Not because I do not want connection — I wrote a whole entry about finding tribe — but because the executive cost of initiating is so high that my system defaults to avoidance unless the context is structured enough to reduce the demand.
I still have reduced responses to social overtures. People reach out and I do not respond — not because I do not care, but because the processing queue is full, or the response requires a type of social performance I cannot generate on demand, or I literally did not register the overture as something that required a response.
I still have inflexibility that is obvious to the casual observer. My wife sees it daily. Anyone who spends enough time around me sees it. The rigidity around routines, the distress when plans change, the difficulty switching between tasks — it is not subtle. It is not hidden. I mask it in brief social encounters, but in sustained contact, it is visible.
These are not Level 1 descriptions. Level 1 says "difficulty initiating." Level 2 says "limited initiation" and "reduced or abnormal responses" and "apparent even with supports in place." The distinction is not about whether you can perform normalcy for an hour in a clinician's office. It is about what your functioning actually looks like across contexts, over time, with all your support systems running.
The Masking Confound
This is where my spike graph entry becomes directly relevant. The compliance officer — the internal system that has been suppressing my actual spikes for decades — is the same system that made me present as Level 1 in a clinical assessment.
A clinician sees you for an hour, maybe two. You are performing. You are running the social scripts at maximum capacity because the context demands it and you have been trained since childhood to rise to the occasion. You make eye contact because you learned to. You modulate your speech because you practiced. You sit in the uncomfortable chair under the fluorescent lights and you do not stim because the compliance officer is working overtime.
And the clinician writes down: Level 1. Requiring support. Mild.
They did not see the crash afterward. They did not see the two days of recovery from that single appointment. They did not see the scripts I prepared before arriving. They did not see that the "appropriate social reciprocity" they observed was a rehearsed performance, not spontaneous engagement. They saw the mask. They assessed the mask. They diagnosed the mask.
The YouTube Realization
And suddenly — suddenly — I understand something that has confused me for years.
I have never been able to do the YouTube creator thing. I have tried. I have wanted to. I watch creators like Auticate and Paige Layle and I think: I have things to say. I have a voice. I have perspectives that connect with people — this log proves it. Why can I not just turn on a camera and talk?
Because it is not just turning on a camera and talking. It is unscripted social performance. It is real-time communication without the buffer of writing, without the ability to edit, without the processing time that text affords. It is managing facial expressions, vocal tone, eye contact with a lens, pacing, energy — all simultaneously, all in real time, all while trying to also have a coherent thought.
For a Level 1, that is hard but doable. Awkward, maybe. Uncomfortable, certainly. But achievable.
For someone whose social communication deficits persist even with supports in place — whose default mode is avoidance of unstructured social performance — it is not a confidence problem. It is not a "just do it" problem. It is a support needs problem. The demand exceeds what my system can generate, even with preparation, even with practice, even with every compensatory strategy I have.
I am not bad at YouTube because I am shy. I am not bad at YouTube because I have not tried hard enough. I am bad at YouTube because the medium requires exactly the type of real-time social performance that represents my highest support need — and I have been judging myself against a severity level that understates that need.
Thirteen years of wondering why I cannot do something that "should" be possible for someone with my intelligence and my ideas. And the answer was in the diagnostic criteria the whole time, just filed under the wrong level.
What This Is Not
I want to be precise about what I am saying and what I am not saying.
I am not saying my original diagnostician was incompetent. The assessment was done with the information available, in a context — a clinical office, a structured interaction — that is structurally biased toward underestimating the support needs of high-masking adults. That is a known limitation of autism assessment, not an individual failure.
I am not claiming to be more disabled than I am. I am claiming to be more supported than the Level 1 designation suggests — and that the support is doing more load-bearing work than anyone, including me until recently, realized.
I am not seeking a different label for its own sake. I am seeking an accurate understanding of my support needs, because accurate understanding is the foundation of effective self-management. If I believe my needs are Level 1 and they are actually Level 2, I will chronically under-accommodate myself and wonder why I keep hitting walls. Which is, not coincidentally, exactly what has been happening.
And I am not catastrophizing. My life is good. My support systems work. I am not in crisis. But the fact that the systems work does not mean the needs are small. It means the engineering is good. And mistaking good engineering for low need is the fundamental error in how severity levels get assigned to high-masking autistic adults.
The AI Mirror
There is something both humbling and clarifying about having an AI synthesize your own data and reflect it back to you without the filters you have been applying to yourself.
I have been writing about my support needs for months. Describing them in detail. Documenting the accommodations, the workarounds, the costs. And somehow, I was still filing all of it under "Level 1, requiring support, basically fine." Because that is what I was told. Because the compliance officer — the same one that flattens my spike graph in clinical settings — was also flattening my self-assessment.
The AI does not have a compliance officer. It just reads the data. All of it. Without the narrative filter that says "but you are high-functioning, so it cannot be that serious." Without the decades of internalized minimization that makes every autistic adult underreport their own needs.
And when you remove that filter, the picture changes.
The Protocol: The severity level you were assigned in a clinician's office may reflect the mask, not the person underneath it. If your support needs are substantial — if the systems holding your life together are doing heavy, structural, daily work — then the fact that life looks functional is evidence of good engineering, not low need. Do not confuse the two. And if an AI reads all your data and says what you have been afraid to say to yourself — sit with it. The compliance officer will try to talk you out of it. The data will not.
