cognitive prosthetics: AI as assistive technology for neurodivergent minds

by Ray Svitla

prosthetics extend capability. glasses correct vision. wheelchairs enable mobility. hearing aids amplify sound.

AI assistants can do the same for executive function.

this isn’t about “life hacks” or “productivity tips.” it’s about assistive technology for people whose brains work differently.

ADHD, autism, dyslexia, and other neurodivergent patterns often come with gaps in executive function: planning, task initiation, working memory, time perception. these gaps are real, frustrating, and exhausting.

AI can fill some of these gaps. not as a cure — nothing about neurodivergence needs “curing” — but as support infrastructure.

the initiation problem

neurotypical person: “I should do the thing” → does the thing

ADHD person: “I should do the thing” → somehow four hours pass → thing still not done

it’s not laziness. it’s not lack of desire. it’s a specific deficit in task initiation.

AI assistants help with this in surprisingly direct ways.

instead of staring at a blank page frozen by overwhelm, you can tell Claude Code: “I need to write a project proposal but I don’t know where to start.”

it gives you structure. a template. specific questions to answer. the activation energy drops from “create something from nothing” to “respond to prompts.”

that difference is huge.

working memory as external storage

ADHD working memory is like RAM that randomly unloads. you have a great idea, you turn to write it down, by the time you open the app you’ve forgotten what the idea was.

AI assistants become external working memory.

you can brain-dump to Claude Code without worrying about structure or completeness. “I’m thinking about X and also Y relates to Z and we should probably consider W.”

it holds the context. it organizes it. it reminds you of the thread you were pulling when you got distracted.

you’re not trying to hold everything in your head while also doing the thing. you offload the holding part.

time blindness: making the invisible visible

ADHD time perception is broken. “this will take 10 minutes” → actually takes 2 hours. or vice versa.

AI can help calibrate.

“I need to write a blog post” → Claude Code: “that typically takes 2-3 hours for drafting plus 30 minutes for editing, want me to block time on your calendar?”

it’s not perfect. but it’s better than “I have no idea how long this takes and will discover that I’m wrong when I miss the deadline.”

also: explicit breakdowns help. “write blog post” is overwhelming. “spend 20 minutes outlining, then 45 minutes on first draft, then 15 minute break” is concrete.

AI can generate these breakdowns automatically based on the task.

the context-switching tax

neurodivergent brains often have high context-switching costs. getting into flow is hard. getting interrupted breaks everything.

AI assistants reduce unnecessary switching.

instead of: → writing code → switch to browser to search API docs → lose context → struggle to resume

you do: → writing code → ask Claude Code about the API → get answer inline → continue writing

one less switch. one less context loss.

same with project management. instead of jumping between code and Linear and Slack, you can ask Claude Code to handle the administrative overhead while you stay in flow.

the paralysis of choice

executive dysfunction often manifests as choice paralysis. too many options → complete freeze.

“should I use React or Vue or Svelte?” → three hours of research, no decision, mounting anxiety.

AI can help: → articulate your constraints → eliminate clearly wrong options → make a recommendation based on your specific situation → give you permission to just pick one

that last part matters. sometimes you don’t need the optimal choice. you need any choice so you can move forward.

hyperfocus: redirecting instead of fighting

ADHD hyperfocus is a double-edged sword. incredible for deep work. terrible for balance.

you meant to spend an hour on the task. six hours later you’re still going, you forgot to eat, you missed two meetings.

AI assistants can gently interrupt hyperfocus:

“you’ve been working on this for 3 hours, which is longer than you planned. take a break? or should I reschedule the 2pm meeting?”

it’s the external time-keeper you can’t be for yourself.

rejection sensitivity: depersonalized feedback

rejection sensitive dysphoria (RSD) is common with ADHD. criticism feels catastrophic, even when it’s mild and constructive.

getting code review feedback from a human: emotional spiral about being a bad developer

getting feedback from Claude Code: “oh, I should handle this edge case, cool”

it’s not that AI feedback is inherently better. it’s that it doesn’t trigger the RSD response.

you can iterate freely without the emotional overhead of feeling judged.

pattern matching: when autism meets AI

autistic cognition often excels at pattern recognition but struggles with implicit social rules.

“how formal should this email be?” is not an autistic-friendly question. there’s no clear algorithm.

AI can help translate:

“I need to email a professor asking for a deadline extension” → Claude generates appropriate tone, explains why certain phrasings work, and gives you confidence the message won’t be misinterpreted.

it’s not masking. it’s translation assistance.

sensory filtering: reducing cognitive noise

processing sensory input is exhausting for many neurodivergent people. too much information, all at once, all equally loud.

AI can filter:

→ summarize that 50-message slack thread → extract action items from the meeting notes → tell me what changed in this code review

you’re still getting the information. but you’re getting signal without noise.

the masking reduction potential

masking — performing neurotypicality — is exhausting and harmful.

AI assistants can reduce the need to mask.

instead of: → pretending you understood the requirements → asking clarifying questions feels like admitting weakness → proceeding with incomplete information → failing because the requirements weren’t clear

you do: → “I didn’t fully understand the requirements, can you help me break this down?” → Claude Code asks clarifying questions you can answer → generates a spec you can verify with the human → no masking required

executive function coaching

ADHD coaching is expensive. AI assistants are not coaches. but they can provide some coaching-adjacent support:

→ breaking down overwhelming tasks → providing structure when you can’t create it yourself → reminding you of your own systems when you forget them → celebrating small wins (dopamine is scarce, take it where you can)

it’s not therapy. it’s scaffolding.

the dangerous side: dependency vs support

here’s the uncomfortable question: does using AI for executive function support make you more capable, or more dependent?

if you can’t function without the AI, have you gained support or have you created a new disability?

I don’t have a clean answer. prosthetics are dependencies too. nobody considers glasses “cheating at vision.”

but there’s a spectrum: → using AI to reduce friction (good) → using AI to develop new capabilities (good) → using AI to avoid building any executive function skills (probably bad?)

the line is blurry and personal.

when AI assistance isn’t enough

AI can help with task initiation, working memory, planning. it can’t help with:

→ emotional regulation → sleep hygiene → medication management → deep trauma work → building sustainable routines

these need humans: therapists, doctors, coaches.

AI is a tool. useful, sometimes essential. but not sufficient.

the accessibility framing

using AI for neurodivergent support should not require justification or apology.

you wouldn’t ask someone why they “need” a wheelchair. don’t ask why someone “needs” AI to help with executive function.

cognitive prosthetics are legitimate assistive technology.

the fact that they’re also useful for neurotypical people doesn’t make them less valid for neurodivergent people.

building AI tools with neurodivergence in mind

most AI assistants are designed for neurotypical workflows. they assume: → consistent motivation → reliable working memory → stable executive function → predictable focus

building for neurodivergent users means designing for: → variable energy and focus → high context-switching costs → need for external structure → working memory gaps → time blindness

different constraints lead to different designs.

the Ivan Illich angle: tools for conviviality

Illich wrote about “tools for conviviality” — technology that enhances human capability without creating dependence on institutions.

AI assistants could be convivial tools. they’re accessible, they enhance capability, they don’t require institutional gatekeeping.

or they could become new forms of institutional control. subscription-only access to cognitive assistance. surveillance-laden executive function support.

the technology enables both futures. which one we get depends on how it’s built and who controls it.

do you use AI as cognitive assistance? what works, what doesn’t? and how do you think about the line between useful support and problematic dependency?

Ray Svitla stay evolving 🐌