AI as Assistive Tech for Autistic Communication

A lot of my energy at work doesn’t go on the work. It goes on the wrapping — the constant, low-level translation between what I mean and what neurotypical communication expects me to perform. Reading tone I can’t see. Softening a direct statement so it doesn’t read as blunt. Decoding an email that says one thing and apparently means another. By mid-afternoon I’m not tired from my job; I’m tired from the interface to my job.

That translation work has a name a lot of us know intimately: masking. And it’s expensive. So I’ve been experimenting with AI assistants as a kind of assistive technology for the communication layer — not to make a better mask, but to spend less of myself maintaining one. There’s an important difference, and I’ll come back to it.

Decoding tone I genuinely can’t see

Some people read subtext automatically. I don’t, reliably. A message like “interesting choice” could be neutral, sarcastic, or a quiet warning, and I will burn real cycles trying to work out which — usually assuming the worst. Pasting it into a chat and asking “what are the possible readings of this, neutral to negative?” gives me a small map instead of a single panicked guess. Sometimes it surfaces a friendly interpretation I’d never have considered on my own.

I hold these readings loosely — more on that below — but having three plausible options is far less destabilising than staring at one ambiguous line with no idea where I stand.

Translating into “work-polite” without losing myself

My natural register is direct. I say what I mean in the fewest words, which is efficient and, to some people, apparently alarming. I’ve spent years manually adding the cushioning — the “just,” the “I was wondering if,” the three sentences of warmth before the actual point — and it never stops feeling like a second language.

Letting an assistant do a first pass of that translation saves me a genuine chunk of energy. I write the blunt, true version, then ask it to make it “warm and collaborative,” then I edit it back toward something that still sounds like me. The key is that last step. The goal isn’t to erase how I communicate — it’s to get past the gatekeeping quickly so I have energy left for the actual conversation.

Prep, scripts, and shrinking the cognitive load

Before a meeting I’m nervous about, I’ll talk through what I want to say and ask for likely questions, so I’m not improvising live — improvising is where masking costs the most. For a long, sprawling email thread, asking for a plain summary of “what’s actually being decided and what’s being asked of me” cuts through the noise so I’m not re-reading the same forty messages to extract three facts. Less load on the parts of my brain that are already working overtime.

The line I won’t cross: scaffold, not mask-maker

Here’s the difference I promised. There’s a version of all this that’s just better masking — using AI to more convincingly perform someone I’m not, so I can keep hiding how I actually work. That version leads to more burnout, not less, because the underlying cost — pretending — is still being paid, just more smoothly.

The version I’m interested in is scaffolding: a ramp that gets me access without demanding I become a different person. A ramp is assistive tech. A more comfortable mask is still a mask. I try to use these tools to conserve energy and reduce friction — and, slowly, alongside that, to ask the people around me to meet me partway, so I need the wrapping less in the first place.

Where it falls short

Its read of tone is a guess, and a confident one. It does not know your colleague; it’s pattern-matching on text. I’ve been handed a “definitely sarcastic” verdict on a message that turned out to be perfectly sincere. Treat its interpretations as options to weigh, never as facts.

Don’t let it talk you out of your boundaries. A tool optimised to sound agreeable will happily smooth a “no” into a “yes, of course” if you let it. The point is to communicate me more easily — not to negotiate me away.

And mind the privacy line. Real names, medical details, anything sensitive about you or anyone else — assume it could be stored. I keep personal specifics out and work with the shape of a message rather than its private contents.

The honest summary

The work was never the expensive part for me. The translation was. Used as a scaffold — and held with healthy suspicion about what it can actually perceive — an AI assistant gives me back some of the energy that masking quietly drains, so there’s more of me left for the things that matter. Not a better disguise. A shorter ramp.