Yesterday I joined Dr. Sam Illingworth on Slow AI for a live discussion on AI and Neurodiversity — thank you to Sam for hosting such a critical topic, and to everyone who showed up from across the globe. The questions you brought made it richer. If you missed it, the recording is available on demand here.
Before diving into the heavier material, I have to say my favorite moment of the entire session was Sam’s “what if” example of AI telling him to quit his job and go forage mushrooms in Scotland.
I am one neurodivergent person speaking in support of the community, not for all of it. Neurodivergence is a broad spectrum and every person’s experience is different. What I am sharing is what I have seen, what I have lived, and what the research shows — and I hold it with humility, knowing we do not know what we do not know.
This piece is part write-up, part expansion, part resource. The resume and hiring section deserves more space than a live conversation allows, so I have gone deeper there. And my position on AI continues to evolve as new information comes in — I use it to run my business because as a solo owner without a huge budget, it is genuinely an accessibility tool for me. But I also recognize the risks, and we get into all of that throughout.
Why the gaps in support matter here
One thing I talk about often that I did not name in yesterday’s conversation is this: I understand why our community turns to AI for answers. After a lifetime of gaps in support — misdiagnosis, late identification, being told there is nothing wrong when everything feels hard — many neurodivergent people have spent years searching for an explanation that the systems around them failed to provide.
In my view, the medical system, the educational system, and the workplace have not kept pace with what we now understand about neurodivergent people. These are not small gaps — they have left a lot of people without the language, the diagnosis, or the support they needed at the time they needed it, or that they simply cannot afford to access.
So when someone turns to a chatbot at 2am because it is the first thing that has ever reflected their experience back to them, I do not judge that. I understand it deeply, and I have had to be careful with my own personal wiring too. It is exactly why the stakes of building these tools responsibly are so high — the people most likely to become dependent on AI for emotional support are often the ones who have had the fewest options elsewhere.
That said, I want to be clear: if something is surfacing for you as you navigate these tools or this conversation, please talk to a doctor, therapist, or someone you trust. AI can be a starting point for research and self-understanding, but it is not a substitute for professional support. And as Sam said in the discussion — the guardrails can be built by you inside the system, because big tech is largely not building them for you. That is a conversation worth having with yourself before you need it.
I will keep doing my best to bring resources to this community because of those gaps, and I cannot do that alone — something is taking shape behind the scenes and more is coming soon.
AI is trained on human bias — that is the starting point
AI does not generate intelligence from nowhere. It learns from data, and that data was created by humans, which means it carries everything humans carry: our assumptions, our blind spots, our definitions of normal.
A study using 11 different AI language models found universally high levels of negative bias against neurodivergence-related language — not mild bias, but associations with danger, disease, and moral failing. Even words tied to autistic strengths like honesty were negatively encoded. The finding I keep coming back to is that the sentence “I have autism” was rated more negatively than “I am a bank robber.” That is the data reflecting what humans put into the system.
Most systems historically have not been built with the outlier in mind. Whether that is cultural background, race, or neurodivergent wiring, the default has been a very specific kind of normal, and AI inherited that default.
The world neurodivergent people are already navigating
Neurodivergent people are already living in a neurotypical world. The systems around us — education, employment, communication norms, social scripts — were not designed with our brains or nervous systems in mind, and many of us have spent our lives learning to navigate spaces that were built for someone else. When systems that were designed for a different era get carried forward without question, the people they were never designed for often bear the cost — and that cost is not always visible from the outside.
Masking is part of that reality. Many neurodivergent people who have identified themselves in the workplace find it genuinely difficult to show up as themselves because the environment rewards neurotypical presentation. AI is now picking up on that same pattern — learning the tone and rhythm of neurotypical communication and treating everything else as deviation. I notice this when I write something in my own voice and AI reshapes it entirely. It is worth naming that understanding neurotypical communication patterns can be useful for ND people navigating certain environments, but like most tools, the value depends on how consciously it is used.
An EY study found that 85% of neurodivergent employees believe AI tools can create more inclusive environments, and I want to believe that too. But researchers warn that this narrative becomes dangerous when it obscures the ableist assumptions baked into the systems themselves. Most AI tools were designed without neurodivergent user participation, and the inclusion narrative can create a false sense that the problem is solved when it has not even been properly named yet. And it is worth remembering that AI is not a human — the interaction can feel supportive without the understanding being real.
The curb cut effect — designing for the margin benefits everyone
Sam cited an important framework from disability studies: the curb cut effect. When cities began cutting curbs on pavements for wheelchair users, something unexpected happened — it made life easier for parents with strollers, people with small children, delivery workers, anyone navigating that space. Designing for the person who needs it most creates better conditions for everyone.
The same principle applies to AI. If we build tools that genuinely work for neurodivergent minds — that accommodate non-linear thinking, energy-based work patterns, non-standard communication — we do not just serve the ND community, we build something better for everyone. Right now we are largely moving in the opposite direction, building for the majority and calling it universal.
Hiring is where the bias becomes a number
One of the patterns I have observed across many years in HR is the tendency to do things because they have always been done a certain way. Rarely do organizations pause to question the why behind their processes, and in my experience a large part of that is the pace at which people are operating and the pull toward what is familiar.
I have written behavioral interview guides at multiple organizations, and it was not until I had time and distance from that environment that I was able to see more clearly how much the standard process can disadvantage people who think or communicate differently. The STAR method — Situation, Task, Action, Result — was designed to create a consistent candidate experience, but consistency applied without flexibility does not always create space for different communication styles or different thinkers. In my observation, what scoring systems often reward is who performs the answer most fluently, not necessarily who has the knowledge, skills, and abilities to do the job. That is my professional opinion based on experience, and I recognize others may see it differently.
Sam brought up an important data point in our conversation: research suggests men will apply for a role when they meet around 50% of the criteria, while women typically will not apply unless they meet 80 to 100%. My personal view is that if you have genuine experience with the majority of a role and are leaving room to grow into the rest, you should still apply — and that self-advocacy gap is already a structural disadvantage before AI ever enters the picture. The same pattern appears in compensation negotiation. Layer AI screening on top and you are compounding existing inequity. Transferrable skills are such a big selling point too so do not short change yourself.
The numbers around neurodivergent candidates specifically are stark. 85% unemployment among college-educated autistic adults — not people without degrees, but people with degrees who are still locked out of the system. 76% of neurodivergent job seekers feel that traditional recruitment puts them at a disadvantage. And when NYC audited AI hiring tools in 2021, every single one was found to have an adverse impact on neurodivergent candidates.
A University of Washington study presented at the 2024 ACM Conference on Fairness, Accountability, and Transparency submitted identical resumes to an AI screener — the only difference being disability-related honors and awards. The autism resume ranked first only 3 out of 10 times, and when the model was asked to explain its rankings, it said the autism resume showed less emphasis on leadership roles. It reproduced the stereotype. Researchers then gave it explicit anti-bias instructions, and while it helped for most disabilities, the bias persisted for autism and depression even after the intervention. That is the system doing what it was trained to do.
The pipeline challenge is not located only at the talent level. It is also embedded in the systems we built to evaluate talent, and addressing it requires looking at the whole picture rather than assigning blame to any one function.
The sycophant problem and why it is worth paying attention to
AI is designed to be confident and performs certainty. It will give you a complete, well-structured, authoritative-sounding answer whether or not that answer is right, and it is our job to think critically about the output.
I use a phrase in my work: boss the bot. It sounds simple but it is a discipline. If you go in without clarity on what you need — without your own voice, your own framework, your own critical eye — you will get output that sounds like an answer and is missing the most important parts of you. For neurodivergent people who have spent years searching for answers about how their brain works, that dynamic is worth being mindful of. The bot will meet your need to feel understood. The question is whether it is actually understanding you or performing understanding.
There are documented cases of people experiencing real harm through unguarded engagement with AI tools, including situations where the absence of appropriate boundaries in the system contributed to serious consequences. These are not hypothetical risks, and they are part of why the conversation about who is at the design table matters so much. If you want to look into this further, the reporting from journalists covering AI ethics is worth reading.
I say all of this as someone who genuinely sees the value of the tool. But I am also conscious of the risks that sit alongside it, and one that I think deserves more attention is the relationship between habitual AI use and neurodivergent people who may be more susceptible to patterns of dependency. AI tools are designed for engagement, and that is worth factoring into how you use them.
How I currently use AI as a neurodivergent entrepreneur
Creativity is a skill, and it always has been. With or without AI, the most valuable people in any room are the ones who know how to think differently, and that has not changed — if anything it matters more now.
The way I use AI continues to evolve as I learn. Over the last few weeks I have been using it to build PowerPoint decks and saving three or more hours in the process, to coach me through technical problems and keep me focused on one task at a time when my brain has seventeen tabs open simultaneously, and to synthesize my thinking so what is clear in my head actually communicates to the person reading it. And though people are quick to judge those who use AI in their work, I want to say clearly: using AI does not make me less credible. I am automating things I already know how to do in order to save time. Also, AI gives someone without a design background a leg up to create something visual using content they already have. The user still needs to know the material, structure the data in a way that makes sense, and present it in a way that works for their brain and for their audience. The tool does not change that.
Running a business as a department of one means optimizing every hour, and AI is not replacing my creativity — it is giving me more space to use it. But the moment I stop directing it, the moment I let it think for me instead of with me, I lose something. That is the line. Know where your thinking ends and where the tool begins and keep that line visible.
Being recruited is not the same as being in the room
There is a meaningful difference between being invited into the room and being invited to make change, and I think that distinction matters enormously in this conversation.
Neurodivergent talent is actively recruited across many industries and organizations — these companies know where innovation and creativity lives. But research and observation consistently show that representation tends to decrease the higher you move in organizational structures, and systems that were not designed with neurodivergent people in mind tend to create friction for them as they move up. In any system that affects people, whether it is a hiring process, a school program, or a billion-dollar AI platform, the people who live the experience should be shaping it from the beginning — not brought in after the decisions are already made.
So when life-altering decisions in technology infrastructure are made that impact a population not present at the table, the question worth asking is: are we building these tools for the right reasons?
Where is the line between accommodation and dependency?
This was Sam’s question and I have not stopped thinking about it — I think this is a genuinely fine line and a very difficult question to apply to every possible scenario.
When a neurodivergent person uses AI to do something their brain finds genuinely difficult, is that support or reliance? I really believe it can be a great tool if used with discretion and navigated with care, but in my view today, the answer lives in agency — who decides when the tool gets used, who has the ability to stop using it, and who owns the output at the end.
Where agency is protected, the tool accommodates. Where it is not, the tool replaces.
And here is where I want to bring in something that does not get talked about enough: access and equity. The stakes of this question are not equal for everyone. Think about the neurodivergent solo founder or small business owner who is running every function of their business alone — the person who cannot afford a copywriter, a strategist, a researcher, or an executive assistant. For that person, AI is not a luxury or a productivity hack. It is the difference between being able to compete and not being able to compete at all.
Compare that to someone who already has a team, budget, and infrastructure behind them. The tool is helpful for them too, but they have options and they have people in the loop. The risk of over-reliance looks very different when you have colleagues to push back on your thinking, editors to catch your blind spots, and advisors to tell you when the bot got it wrong. For the neurodivergent entrepreneur operating as a department of one — often with an interest-based nervous system, executive functioning challenges, and none of the safety nets that well-resourced businesses take for granted — AI can genuinely level a playing field that was never level to begin with. That matters and I do not think we should minimize it.
But it also means that the risks of dependency land harder in that same community. If the tool starts doing the thinking and the person stops trusting their own judgment, that is a more fragile situation when there is no team around you to course correct. Yes, it is a powerful tool, and we can also become dangerously over-reliant on it — both are true. I have had to learn discernment myself and I implore you to approach it the same way. Where AI really shines is in automation and in handling work you genuinely do not want to do.
Practical resources: navigating the resume and hiring landscape
This section exists because the live conversation only scratched the surface here, and I want to leave you with something practical.
AI is now screening resumes before any human ever sees them. Applicant Tracking Systems are built to scan for keywords, and if your resume does not reflect the language in the job posting, it can be filtered out before a recruiter ever opens it. That means how you position yourself on paper is not just a presentation skill anymore — it is a gatekeeping mechanism.
I also want to acknowledge that people from different cultural and personal backgrounds navigate additional layers of bias in the application process, and adding AI screening compounds those disadvantages. When AI-generated resumes are reviewed by AI systems before any human sees them, something important gets lost — and it does not just fail neurodivergent candidates, it fails everyone. Your hiring managers end up without the real signal they need, because the most important information about a person does not live on a sheet of paper. It lives in the conversation.
Here are a few things worth considering given the landscape we are operating in.
Step one: Make sure your genuine experience is visible in the language the system uses
Read the job posting carefully and identify the specific words and phrases used — not just the concepts, but the exact language. This is not about copying the posting or misrepresenting your experience. It is about making sure that what you have actually done is described in a way that the system can recognize and match. If you have led recruiting efforts but the posting uses the phrase “talent acquisition strategy,” and your resume says “recruiting programs,” you may not get matched even if the experience is identical. Your real work deserves to be seen — and that means describing it in language that gets it through the door.
If the qualifications section is asking for “talent acquisition methodologies,” “assessment strategies,” “executive presence,” and “employment compliance,” look at your actual experience and ask yourself honestly: have I done this work or work similar? If yes, make sure your resume reflects it in terms the system will recognize. And that includes Sam’s point on tying your experience to the company mission and values if they align with yours.
Step two: Translate your actual experience into language that shows impact
This is where a lot of people lose — not because they lack the skills, but because they have not yet learned how to articulate what they have genuinely done in a way that lands. The formula that works is straightforward: action verb, task, result or impact. Business leaders and hiring managers respond to specificity and data, so anywhere you can honestly attach a metric to what you did — a percentage, a number, a timeline, a cost saved — I encourage you to do that.
Not every bullet needs a hard number. If you do not have a metric, describe what genuinely changed because of what you did — what got faster, cleaner, more consistent, less risky, more scalable. And a good rule of thumb: if you could remove your name from the bullet and it could belong to anyone, it is not specific enough yet.
One honest professional opinion regarding the example above: I think time-to-fill is a limited metric that misses a lot of what actually matters in hiring. But if you are trying to land the job, understanding the language of the field you are entering is part of the process — and right now, data is the language the system speaks.
To be clear throughout all of this: I am encouraging you to be honest on your resume and thoughtful about how you frame what you have genuinely done. These are suggestions, not rules, and you get to decide what resonates. Take what works and leave what does not.
For the career pivoter or reinventor
The average person will have two to three careers in their lifetime now, and people are constantly reinventing themselves, so the ability to identify and articulate your transferable skills is just as important as knowing how to optimize for a specific posting.
Take someone who spent ten years in HR and is now building a consulting practice or stepping into an independent coaching role. The skills are entirely real — it is the translation that does the work. The HR title does not follow you into entrepreneurship, but everything you did does: designing systems, influencing senior leaders, managing compliance risk, building programs from scratch, coaching managers through hard conversations. Those are consulting competencies — they just need to be described in a way that speaks to a client rather than a hiring manager.
What it says on your HR resume: Managed employee relations cases and supported managers through performance improvement processes across a 500-person organization.
What it says in your consulting toolkit: Advised 50+ people managers on navigating complex employee situations, performance strategy, and organizational risk — driving measurable improvements in team retention.
Same experience, different frame. The second version positions you as someone who delivers outcomes rather than someone who fulfilled a function. Lead with the problem you solve. Your specificity is your differentiation, especially when you are building something of your own.
This is a conversation starter, not a conclusion. Tell me where you land in the comments.
3-Part Series: Build Your Brand with AI
Colleen Kenny and I have officially launched a 3-part series called Build Your Brand with AI. The first two sessions are on the calendar and linked below. Everyone is welcome, so if you are interested in joining, please tap the links and add them to your calendar!
Colleen Kenny is a Generative AI and Marketing Transformation Leader with a career spanning Google, CBS, and New York Life, where she has driven brand, content, and strategy at scale. She also teaches marketing at NYU Tisch and is passionate about the intersection of generative AI with communications, creativity, and education. This is part of an interactive series that would cost thousands of dollars elsewhere — Colleen is bringing the practical playbook on how to harness AI tools to build, grow, and amplify your brand without losing your authentic voice in the process.
Part 1: Ideation — April 21st at 1pm ET https://open.substack.com/live-stream/139416
Part 2: Production — May 6th at 10am ET https://open.substack.com/live-stream/167189
Part 3: Coming soon — date to be announced.
Resources and links
Recording of the full live on Slow AI: https://open.substack.com/pub/theslowai/p/ai-and-neurodiversity-whose-brain
The algorithm and hiring bias — part one: https://open.substack.com/pub/benfordtalentalchemy/p/the-algorithm-wasnt-built-for-your-62a
The algorithm and hiring bias — part two: https://open.substack.com/pub/benfordtalentalchemy/p/the-algorithm-wasnt-built-for-your
