Reactive Machines

When your brain works differently, AI isn’t a luxury—it’s accessibility

AI as accessibility: what happened when a neurodivergent solutions architect stopped fighting his brain and started building.

In this post, I share how AI serves as an accessibility tool for neurodivergent professionals. The system is built on Amazon Quick on your desktop, an AI-powered desktop and web assistant that compensates for executive function gaps every day. I also share questions to help you identify where AI could support yours. Approximately 15–20 percent of the UK adult population is neurodivergent, according to research from Birkbeck, University of London. Most AI productivity tooling still assumes neurotypical brains. Email triage, prioritization, and follow-up management consume disproportionate cognitive energy for neurodivergent professionals compared to the technical work itself.

I have AuDHD (co-occurring autism and ADHD). My brain is excellent at pattern recognition, deep analytical thinking, and creative problem-solving. It’s terrible at remembering what happened yesterday, deciding what to do next, switching between tasks without losing context, and maintaining organizational systems.

For years, I compensated. I masked. I built elaborate workarounds. And every evening, I came home depleted, having spent all my executive function at work, with nothing left for the people who matter most.

Then I started building with AI. Not “asking ChatGPT to write email” but actually building systems that could compensate for the specific cognitive gaps my neurodivergence creates.

This is what happened.

What executive function actually costs

Most people don’t understand this about AuDHD: the two conditions are in constant conflict. My autistic brain craves structure, routine, and predictability. It wants the perfect system. My ADHD brain resists routine, craves novelty, and cannot sustain any system once the initial dopamine wears off. They’re permanently at war.

So I build the beautiful system (autism satisfied), use it enthusiastically for a week (ADHD riding the novelty hit), then abandon it completely when it becomes effortful (ADHD wins). Then I feel genuine distress at the resulting chaos (autism screaming). Repeat forever.

Organizing costs me 10 times the cognitive energy a neurotypical brain expends. Even when I pay that cost, the system I built won’t survive contact with my own neurology.

The “tool graveyard” cycle isn’t laziness. It’s two competing neurotypes making it impossible to sustain any organizational system.

Asana, Notion, Todoist, paper planners, whiteboards. I’ve tried them all. They all died.

The insight that changed everything: building a system that maintains itself.

Building the invisible scaffold

Over the past several months, I’ve built an AI-powered workflow system. It runs alongside my work. I start it each morning. Security controls mean it’s not fully autonomous. Once running, it handles the thinking. It observes, classifies, acts, and reports. My only job is to start it.

What it does every morning:

Email triage: One action to start, zero thinking required. After I start the scan, the system classifies every message against refined rules. It then presents me with a prioritized briefing. It identifies direct asks and time-sensitive messages. It also flags noise I can safely ignore.

For an AuDHD brain, inbox is death. The ADHD sees 50 unread email messages as a paralyzing wall of undifferentiated demand. The autism sees 50 email messages requiring correct processing in the right order. If I can’t determine the right order, I can’t start at all.

My system turns it into “3 things need you today, 2 are waiting on others, everything else is handled.” Both neurotypes can accept that.

Task state management and priority decisions work the same way, running as a single integrated layer beneath my daily work.

I don’t manually move a task, and I no longer forget to follow up. The system does what my working memory cannot.

Priority is where AuDHD creates a nightmare from both sides. The ADHD means everything feels equally urgent (or equally unimportant), with no internal ranking signal. The autism means that I will over-research a low-priority task for 3 hours because I need to understand it completely before I can act. Meanwhile, the actually urgent thing goes ignored.

I built explicit rules into my system. “Do First” means someone is actively waiting on me, I can act right now, and it’s time-bound. If any of those conditions aren’t true, it automatically demotes.

The system alleviates the 20-minute paralysis over what to do next. The answer’s already there.

What I actually built

Figure 1. The design of the AI tooling for email triage

This is more than prompting an AI and copying the output. The system runs on the Quick desktop application, which provides the persistent AI assistant, conversation memory, and tool orchestration layer. Quick connects to Amazon Bedrock for its underlying inference. This means the AI reasoning adapts as foundation models improve without requiring changes to my workflow.

The custom piece is a Model Context Protocol (MCP) server. I built it using Kiro, an AI-powered integrated development environment (IDE) from AWS. This server connects Quick to my Outlook inbox, calendar, and Asana task board. It encodes my triage rules and priority logic as configurable markdown files. Communication patterns are also stored this way. The AI reads these fresh each session. When I refine a rule, the system’s behavior changes immediately. No redeployment required.

The Quick skills framework provides reusable automation patterns for recurring workflows, including email formatting, context logging, end-of-day summaries, and task state management. Each skill runs deterministically when triggered, reducing the cognitive overhead to near zero.

The critical design principle: the cognitive cost of using it is near zero. I start a session, authenticate, and then the system handles everything from there. No daily reviews to maintain. No checklists to complete. No decisions about what to process first. The initiation takes seconds. It’s the thinking the system offloads.

The neurospicy difference

Most AI productivity content is written by people whose brains already work well for office work. They’re using AI to go from good to great.

For many neurodivergent people, AI can be the difference between functioning and not functioning.

The first thing that changed was time blindness. I genuinely can’t feel time passing (that’s the ADHD). But the autism means when I do realize something’s overdue, the shame and social anxiety about the lateness makes it even harder to act. My system tracks elapsed time on every task and surfaces things before they go stale. I never get to the shame stage because follow-ups happen automatically within a week.

Decision paralysis and context-switching tax both dissolve once the system is in place. The ADHD can’t rank importance. The autism can’t start a task without understanding the full context first. Together they produce complete inertia. But when I look at my board in the morning, everything’s already sorted: Do First. Do Next. If Time. I don’t need to decide. I start at the top. Similarly, ADHD causes me to drop context the moment I switch tasks, whereas autism means I need complete context before re-engaging. Every conversation with my AI assistant carries the full context of what we’ve discussed, what’s in progress, and what’s waiting. I can leave a task, come back 3 days later, and the system gives me exactly what my autistic brain needs to re-enter the problem space.

Then there’s the masking cost. Autistic communication is direct. I don’t naturally do the social softening that professional email requires. But I’ve learned that being too direct reads as rude, so I mask. That masking is exhausting. I’ve trained the system on my actual communication style, the version that’s authentically me but calibrated enough for professional contexts. Drafts come out sounding like me, not like corporate AI, and I spend less energy on the performance of professionalism.

The tool graveyard problem also dissolved. The key difference from every system that died before: the initiation cost is minimal. I open it, authenticate, run a scan, and everything else happens without me.

The numbers

All following figures are from my personal experience only. They aren’t product benchmarks or guarantees.

  • Longest sustained workflow streak I’ve ever maintained, by a factor of four. Every previous organizational system died within 10 days.
  • 6–13 minutes for a full inbox scan that previously took me over 45 minutes of manual scrolling and decision-making (estimated from personal experience before the system existed).
  • Zero dropped follow-ups in the past month. Previously I lost an estimated 2–3 per week based on partner feedback and missed chase patterns in my inbox.
  • Consistent end-of-day clarity. I know what I did, what moved, and what’s waiting. That never happened before.

What this means if you’re like me

You don’t need to be a better engineer than anyone else.

If you’re neurodivergent and struggling with the organizational demands of knowledge work, AI isn’t about making you more productive in the neurotypical sense. It’s about making the invisible labor visible and then automating it away.

Your brain is probably excellent at the actual work. The problem is everything around it: the admin, the follow-ups, the prioritization, the remembering.

Start here: Map your three most expensive cognitive tasks, the ones that drain you disproportionately to their value. For each one, ask whether it requires you to remember, decide, or initiate something. If the answer to any of those is yes, that task is a candidate for AI offloading. Then determine whether you can reduce it to a single action. Not “fully automatic” (that’s often unrealistic with security and authentication requirements). But could the thinking be offloaded so all that’s left is starting it? When even one passes that test, you’ve found your first build.

The mapping exercise looks like this:

Cognitive task Remember? Decide? Initiate? Single-action candidate?
Email triage Who’s waiting on me, what’s stale What’s urgent vs. noise Open inbox and start processing ✓ “Run scan” – rules do the rest
Follow-up tracking What I promised, when, to whom When to chase vs. wait Actually sending the chase ✓ Automatic promotion after N days silent
Priority setting Full picture of everything in flight What matters most right now Committing to the next action ✓ Pre-computed board every morning

The questions that unlock it:

  1. Can I write down my decision rule honestly? Write down how it actually works, not how you wish it worked. Urgent isn’t a feeling. For me it means: Someone external is waiting, I can act right now, and there’s a deadline within 48 hours. If any of those conditions aren’t true, it’s not urgent. That’s six lines of markdown. The AI reads it and applies it to every email without fatigue.
  2. Can I separate the thinking from the doing? The act of replying to an email takes 2 minutes. The act of deciding which email to reply to takes 20 minutes of paralysis. Those are different problems. I offloaded the second one entirely.
  3. Can I make the initiation cost essentially zero? If it requires me to remember to use it, my AuDHD will kill it. The threshold is: Can I start it in under 5 seconds, before the resistance kicks in? If yes, it survives. If no, it joins the tool graveyard regardless of how good it is.

The rules themselves are only markdown. Plain English, no programming. Something like:

“Do First” means someone external is actively waiting AND I can act right now AND it’s time-bound. If any of those conditions aren’t true, automatically demote.

That’s the entire priority rule. Edit the file, and the behavior changes next session. No redeployment. No learning curve. Just your own decision logic, written down once, applied consistently forever.

Start with the one that causes the most shame. For me it was dropped follow-ups. The thing I kept apologizing for. The thing that made me feel like I was failing at basic professionalism. That’s your first build. It’s where the cognitive tax is highest and the relief is most immediate.

A note on shame

The hardest part of this journey wasn’t the technical build. It was admitting I needed it.

There’s a voice that says “everyone else manages this without a robot assistant.” That voice is wrong, and even if it weren’t, it doesn’t matter. If you need glasses to see, you wear glasses. If you need a ramp to access a building, the building should have a ramp.

If you need AI to manage your executive function, that’s accessibility. Full stop.

To explore the possibilities for your own workflow, see Amazon Quick and Amazon Quick on your desktop.


About the author

Andrew Johnston

Andrew Johnston

Andrew is a partner solutions architect at AWS focused on UK Public Sector initiatives. With over 30 years in IT across the UK and US, he brings deep expertise from his work with global integrators, SMEs, and software companies. Andrew excels at connecting business needs with technical solutions, helping AWS partners deliver innovative outcomes for their customers. His unique ability to spot patterns and craft elegant solutions makes him an asset in driving public sector digital transformation.

Source link

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button