AI as accessibility: what occurred when a neurodivergent options architect stopped preventing his mind and began constructing.
On this put up, I share how AI serves as an accessibility device for neurodivergent professionals. The system is constructed on Amazon Fast in your desktop, an AI-powered desktop and net assistant that compensates for government perform gaps every single day. I additionally share questions that can assist you establish the place AI might help yours. Roughly 15–20 % of the UK grownup inhabitants is neurodivergent, in keeping with analysis from Birkbeck, College of London. Most AI productiveness tooling nonetheless assumes neurotypical brains. E mail triage, prioritization, and follow-up administration eat disproportionate cognitive vitality for neurodivergent professionals in comparison with the technical work itself.
I’ve AuDHD (co-occurring autism and ADHD). My mind is great at sample recognition, deep analytical considering, and inventive problem-solving. It’s horrible at remembering what occurred yesterday, deciding what to do subsequent, switching between duties with out shedding context, and sustaining organizational programs.
For years, I compensated. I masked. I constructed elaborate workarounds. And each night, I got here dwelling depleted, having spent all my government perform at work, with nothing left for the individuals who matter most.
Then I began constructing with AI. Not “asking ChatGPT to jot down electronic mail” however really constructing programs that would compensate for the precise cognitive gaps my neurodivergence creates.
That is what occurred.
What government perform really prices
Most individuals don’t perceive this about AuDHD: the 2 situations are in fixed battle. My autistic mind craves construction, routine, and predictability. It needs the right system. My ADHD mind resists routine, craves novelty, and can’t maintain any system as soon as the preliminary dopamine wears off. They’re completely at struggle.
So I construct the gorgeous system (autism happy), use it enthusiastically for every week (ADHD using the novelty hit), then abandon it fully when it turns into effortful (ADHD wins). Then I really feel real misery on the ensuing chaos (autism screaming). Repeat without end.
Organizing prices me 10 occasions the cognitive vitality a neurotypical mind expends. Even after I pay that price, the system I constructed gained’t survive contact with my very own neurology.
The “device graveyard” cycle isn’t laziness. It’s two competing neurotypes making it unimaginable to maintain any organizational system.
Asana, Notion, Todoist, paper planners, whiteboards. I’ve tried all of them. All of them died.
The perception that modified the whole lot: constructing a system that maintains itself.
Constructing the invisible scaffold
Over the previous a number of months, I’ve constructed an AI-powered workflow system. It runs alongside my work. I begin it every morning. Safety controls imply it’s not absolutely autonomous. As soon as operating, it handles the considering. It observes, classifies, acts, and stories. My solely job is to begin it.
What it does each morning:
E mail triage: One motion to begin, zero considering required. After I begin the scan, the system classifies each message in opposition to refined guidelines. It then presents me with a prioritized briefing. It identifies direct asks and time-sensitive messages. It additionally flags noise I can safely ignore.
For an AuDHD mind, inbox is loss of life. The ADHD sees 50 unread electronic mail messages as a paralyzing wall of undifferentiated demand. The autism sees 50 electronic mail messages requiring right processing in the precise order. If I can’t decide the precise order, I can’t begin in any respect.
My system turns it into “3 issues want you at the moment, 2 are ready on others, the whole lot else is dealt with.” Each neurotypes can settle for that.
Process state administration and precedence choices work the identical method, operating as a single built-in layer beneath my each day work.
I don’t manually transfer a process, and I now not overlook to observe up. The system does what my working reminiscence can’t.
Precedence is the place AuDHD creates a nightmare from either side. The ADHD means the whole lot feels equally pressing (or equally unimportant), with no inner rating sign. The autism implies that I’ll over-research a low-priority process for 3 hours as a result of I want to grasp it fully earlier than I can act. In the meantime, the really pressing factor goes ignored.
I constructed specific guidelines into my system. “Do First” means somebody is actively ready on me, I can act proper now, and it’s time-bound. If any of these situations aren’t true, it robotically demotes.
The system alleviates the 20-minute paralysis over what to do subsequent. The reply’s already there.
What I really constructed
Determine 1. The design of the AI tooling for electronic mail triage
That is greater than prompting an AI and copying the output. The system runs on the Fast desktop utility, which offers the persistent AI assistant, dialog reminiscence, and gear orchestration layer. Fast connects to Amazon Bedrock for its underlying inference. This implies the AI reasoning adapts as basis fashions enhance with out requiring adjustments to my workflow.
The customized piece is a Mannequin Context Protocol (MCP) server. I constructed it utilizing Kiro, an AI-powered built-in growth surroundings (IDE) from AWS. This server connects Fast to my Outlook inbox, calendar, and Asana process board. It encodes my triage guidelines and precedence logic as configurable markdown information. Communication patterns are additionally saved this manner. The AI reads these contemporary every session. Once I refine a rule, the system’s habits adjustments instantly. No redeployment required.
The Fast expertise framework offers reusable automation patterns for recurring workflows, together with electronic mail formatting, context logging, end-of-day summaries, and process state administration. Every talent runs deterministically when triggered, decreasing the cognitive overhead to close zero.
The vital design precept: the cognitive price of utilizing it’s close to zero. I begin a session, authenticate, after which the system handles the whole lot from there. No each day opinions to keep up. No checklists to finish. No choices about what to course of first. The initiation takes seconds. It’s the considering the system offloads.
The neurospicy distinction
Most AI productiveness content material is written by folks whose brains already work properly for workplace work. They’re utilizing AI to go from good to nice.
For a lot of neurodivergent folks, AI may be the distinction between functioning and never functioning.
The very first thing that modified was time blindness. I genuinely can’t really feel time passing (that’s the ADHD). However the autism means after I do notice one thing’s overdue, the disgrace and social nervousness concerning the lateness makes it even tougher to behave. My system tracks elapsed time on each process and surfaces issues earlier than they go stale. I by no means get to the disgrace stage as a result of follow-ups occur robotically inside every week.
Determination paralysis and context-switching tax each dissolve as soon as the system is in place. The ADHD can’t rank significance. The autism can’t begin a process with out understanding the complete context first. Collectively they produce full inertia. However after I take a look at my board within the morning, the whole lot’s already sorted: Do First. Do Subsequent. If Time. I don’t must resolve. I begin on the prime. Equally, ADHD causes me to drop context the second I swap duties, whereas autism means I want full context earlier than re-engaging. Each dialog with my AI assistant carries the complete context of what we’ve mentioned, what’s in progress, and what’s ready. I can depart a process, come again 3 days later, and the system offers me precisely what my autistic mind must re-enter the issue area.
Then there’s the masking price. Autistic communication is direct. I don’t naturally do the social softening that skilled electronic mail requires. However I’ve realized that being too direct reads as impolite, so I masks. That masking is exhausting. I’ve educated the system on my precise communication fashion, the model that’s authentically me however calibrated sufficient for skilled contexts. Drafts come out sounding like me, not like company AI, and I spend much less vitality on the efficiency of professionalism.
The device graveyard downside additionally dissolved. The important thing distinction from each system that died earlier than: the initiation price is minimal. I open it, authenticate, run a scan, and the whole lot else occurs with out me.
The numbers
All following figures are from my private expertise solely. They aren’t product benchmarks or ensures.
- Longest sustained workflow streak I’ve ever maintained, by an element of 4. Each earlier organizational system died inside 10 days.
- 6–13 minutes for a full inbox scan that beforehand took me over 45 minutes of guide scrolling and decision-making (estimated from private expertise earlier than the system existed).
- Zero dropped follow-ups prior to now month. Beforehand I misplaced an estimated 2–3 per week primarily based on associate suggestions and missed chase patterns in my inbox.
- Constant end-of-day readability. I do know what I did, what moved, and what’s ready. That by no means occurred earlier than.
What this implies if you happen to’re like me
You don’t must be a greater engineer than anybody else.
When you’re neurodivergent and combating the organizational calls for of information work, AI isn’t about making you extra productive within the neurotypical sense. It’s about making the invisible labor seen after which automating it away.
Your mind might be wonderful on the precise work. The issue is the whole lot round it: the admin, the follow-ups, the prioritization, the remembering.
Begin right here: Map your three costliest cognitive duties, those that drain you disproportionately to their worth. For every one, ask whether or not it requires you to recollect, resolve, or provoke one thing. If the reply to any of these is sure, that process is a candidate for AI offloading. Then decide whether or not you’ll be able to scale back it to a single motion. Not “absolutely computerized” (that’s usually unrealistic with safety and authentication necessities). However might the considering be offloaded so all that’s left is beginning it? When even one passes that check, you’ve discovered your first construct.
The mapping train appears to be like like this:
| Cognitive process | Keep in mind? | Resolve? | Provoke? | Single-action candidate? |
| E mail triage | Who’s ready on me, what’s stale | What’s pressing vs. noise | Open inbox and begin processing | ✓ “Run scan” – guidelines do the remainder |
| Comply with-up monitoring | What I promised, when, to whom | When to chase vs. wait | Truly sending the chase | ✓ Computerized promotion after N days silent |
| Precedence setting | Full image of the whole lot in flight | What issues most proper now | Committing to the subsequent motion | ✓ Pre-computed board each morning |
The questions that unlock it:
- Can I write down my determination rule actually? Write down the way it really works, not how you would like it labored. Pressing isn’t a sense. For me it means: Somebody exterior is ready, I can act proper now, and there’s a deadline inside 48 hours. If any of these situations aren’t true, it’s not pressing. That’s six strains of markdown. The AI reads it and applies it to each electronic mail with out fatigue.
- Can I separate the considering from the doing? The act of replying to an electronic mail takes 2 minutes. The act of deciding which electronic mail to answer to takes 20 minutes of paralysis. These are totally different issues. I offloaded the second fully.
- Can I make the initiation price primarily zero? If it requires me to recollect to make use of it, my AuDHD will kill it. The brink is: Can I begin it in underneath 5 seconds, earlier than the resistance kicks in? If sure, it survives. If no, it joins the device graveyard no matter how good it’s.
The principles themselves are solely markdown. Plain English, no programming. One thing like:
“Do First” means somebody exterior is actively ready AND I can act proper now AND it’s time-bound. If any of these situations aren’t true, robotically demote.
That’s your entire precedence rule. Edit the file, and the habits adjustments subsequent session. No redeployment. No studying curve. Simply your individual determination logic, written down as soon as, utilized persistently without end.
Begin with the one which causes essentially the most disgrace. For me it was dropped follow-ups. The factor I stored apologizing for. The factor that made me really feel like I used to be failing at fundamental professionalism. That’s your first construct. It’s the place the cognitive tax is highest and the aid is most fast.
A observe on disgrace
The toughest a part of this journey wasn’t the technical construct. It was admitting I wanted it.
There’s a voice that claims “everybody else manages this with out a robotic assistant.” That voice is mistaken, and even when it weren’t, it doesn’t matter. When you want glasses to see, you put on glasses. When you want a ramp to entry a constructing, the constructing ought to have a ramp.
When you want AI to handle your government perform, that’s accessibility. Full cease.
To discover the chances to your personal workflow, see Amazon Fast and Amazon Fast in your desktop.
In regards to the writer
