Wednesday, May 13, 2026 Intelligence for Portfolio Executives closing the AI Wage Gap.

Earlier this week, I sat down with Ludmila Praslova for Closing the AI Wage Gap.

She's the I-O psychologist behind The Canary Code — the book that named, in plain language, what most workplaces have spent thirty years pretending wasn't happening. The 2025 Thinkers50 Talent Award. The first person to publish in HBR from an autistic perspective. The reason a generation of HR leaders finally have a framework for inclusion that isn't sensitivity training in a different font.

She also features me in her book — as an example of neurodivergent talent navigating the corporate ladder. Three CHRO seats, a J.D. from Cardozo, 2,300+ coaching clients, a Top 5 Global HR Thought Leader designation. And undiagnosed ADHD running underneath the whole thing for forty years, which I only named in the last few.

I expected the interview to confirm my thesis. It confirmed half of it. The other half it broke open in a way I haven't stopped thinking about since.

Here is what she said, paraphrased to one sentence:

AI is the best accommodation neurodivergent professionals have ever had — and the single biggest threat to their creative edge. Both. At the same time.

The first half of that is the easy half. It's the half every neurodivergent professional in your network can already feel in their bones.

The second half is the half nobody is writing about. And it's the half that will quietly hollow out the careers of the people I'm writing this book for, if they don't see it coming.

Start with what's working — because it is genuinely working, and the data is no longer ambiguous.

A UK study circulating in the I-O literature found that neurodivergent workers reported roughly 25% higher satisfaction with AI assistants than their neurotypical peers. The mechanism isn't mysterious. AI removes three invisible taxes that neurodivergent professionals have paid every working day for their entire careers.

The first is the social-tax of synchronous communication. For the autistic professional, every meeting is a translation exercise running in the background while the actual content is being processed in the foreground. For the ADHD professional, every meeting is a regulation exercise — keeping the body still, the mouth shut, the mind from sprinting three topics ahead. The energy cost is real and it does not show up on anyone's productivity dashboard. An LLM removes that tax entirely. You can think at your own speed, in your own structure, without the social maintenance overhead. Alon Bochman — fractional CTO, ex-Google, ex-FactSet, recurring voice in my book — told me directly that talking to Claude is a different emotional experience than talking to a person. Less performance. More signal.

The second is the translation tax. Most neurodivergent professionals have built, over decades, an exhausting bilingualism: their native cognitive language plus the corporate dialect they have to translate it into to be heard. The bilingualism is the reason they often appear less competent than they are. The LLM is the first tool that translates in both directions for free. You can write the way you think, and the model can render it in the register the audience expects. That's not a productivity gain. That's a removal of a tax that compounded over careers.

The third is the executive-function tax. ADHD brains in particular spend extraordinary energy on the meta-work — task initiation, working memory, prioritization, transition costs between contexts. An LLM is a prosthetic for exactly that meta-work. Drop a brain dump into the context window. Ask it to structure. Ask it to remind you what you said yesterday about the thing you forgot you said. Ask it to write the first sentence so you can stop staring at the blank page. The tool is doing the thing the medication has been doing for the people who could access medication — and it's doing it for free, at scale, for everyone.

If you are neurodivergent and you have not yet noticed that the last two years have been quietly the most productive of your professional life, look more carefully.

This is the half that's already true. This is the half that vindicates the thesis I've been writing for eighteen months: that the same brains that corporate ladders broke are the brains best suited to AI-leveraged portfolio careers. The accommodations are real. The wage premium is real. The compounding is real.

And then Praslova said the other thing.

Quote, from the interview: AI output tends to go toward the average, and erases the most creative things neurodivergent brains produce.

Read that twice.

Here is the mechanism, the way she explained it and the way I now understand it.

Every large language model is, by mathematical construction, a regression-to-the-mean machine. It is trained on the distribution of human writing and reasoning, and at every token it samples toward the center of that distribution. The defaults are tuned to be agreeable, structured, on-topic, safe. The model is not erasing variance because it has bad values. It is erasing variance because that is what the loss function rewards.

Now ask: what is the actual economic value of a neurodivergent professional in a knowledge economy?

It is not consistency. It is not reliability under standardized conditions. It is not the ability to produce the median version of the document.

It is variance. It is the unexpected connection. It is the autistic professional's pattern recognition across domains that nobody else thought to bridge. It is the ADHD professional's hyperfocus burst that compresses three weeks of work into a Saturday afternoon. It is the dyslexic professional's spatial reasoning that lets them see the system other people are still reading about. It is, in the language of Praslova's Canary Code, the cognitive diversity that makes the team smarter than any of its members.

That variance is the asset. It is the thing AI cannot replicate, because the model is trained to do the opposite.

And here is the trap: if you, as a neurodivergent professional, accept the LLM's first draft — which feels accommodating, which feels like it understands you, which removes the three taxes I just described — you are accepting the average version of your thinking. Not someone else's average. Your own variance, regressed to the mean.

The accommodation that removes your social tax also smooths out the cognitive signature that was your only durable competitive advantage. You will feel more productive. You will produce more documents. You will be more comfortable in more meetings. And your work, over time, will look more and more like everyone else's.

The canary did not die because the air got dramatically worse. The canary died because the air got slightly worse, continuously, in a way the miners couldn't feel.

This is not a reason to use AI less. That is the wrong conclusion and it would leave the wage gap intact.

This is a reason to use AI with a specific discipline that almost nobody is currently practicing.

Three things. Concrete.

One. Stop accepting first drafts. The first draft is the regression. The first draft is the average. The first draft removed the social tax but it also removed the part of you that the market is actually paying for. Use the first draft as a starting position. Then push back. Tell the model what it missed. Tell it which connection it didn't make. Tell it which framing it defaulted to and which one your brain actually wanted. The second draft is closer. The fourth draft is where your variance lives.

Two. Anchor the model in your specific cognitive signature. Most professionals are using LLMs with no context other than the prompt. That is the worst possible configuration for a neurodivergent user, because it forces the model into maximum default. Build a context document — your style, your recurring frameworks, your provocations, your no-go territory, your way of opening and closing a piece. Feed it to the model every time. Force the model to write like you, not like the average of everyone. This is not optimization. This is anti-erasure.

Three. Build, don't license. This is the structural move. When you use someone else's LLM application, you are accepting their defaults — their tone, their structure, their averaging. When you build your own — even a small one, even a Flask app that fits one workflow — you control the defaults. You set the system prompt. You decide what the model should refuse to flatten. The build-layer is the only place where the cognitive variance of the operator gets preserved end-to-end. This is why the Portfolio Executive OS treats shipping at least one custom AI tool per quarter as non-negotiable. It is not about the tool. It is about owning the layer where the averaging happens.

If I had known, at twenty-five, that the brain that made every standardized environment hard was the same brain that would build six AI products by forty-three, I would have saved myself a decade of trying to fit it.

If I had known, at thirty-five, that the workplaces that broke me would be replaced inside ten years by a career structure that runs on the brain that broke in them, I would have left those workplaces five years earlier than I did.

And if I had known, at forty-three, that the very tool that finally accommodates the way my brain works is also the tool that will quietly average it away if I'm not paying attention — I would have started paying attention sooner.

This is what the book is about. This is what Praslova's interview confirmed and broke open. And this is the move I want every neurodivergent operator on this list to make this week.

Don't just use the tool.

Make the tool work for your variance, not against it.

Closing the AI Wage Gap devotes a structural thread to neurodivergent operators across all five Parts. Not a sidebar. Not a chapter. A thread — because Praslova's own pushback in our interview was that the term neurodivergent is too broad to be operationally useful, and the right response is profile-specific guidance throughout, not a single ghettoized section.

The chapters where the thread is densest:

**The manuscript is in Tier-1 agent querying right now. Levine, Halpern, Sagalyn pending. If you know an editor or agent who works at the intersection of work, AI, and cognitive diversity — reply to this email.

Three platforms paying domain experts to train the next generation of AI models. Updated this edition.

Mercor — Now valued at $10B (October 2025 Series C, $350M led by Felicis). Roughly 30,000+ contractors hired in 2025 alone for projects at OpenAI, Anthropic, and other frontier labs. Average expert rate ~$85/hour; engineering work $70–$200+/hr. Note: Mercor was impacted by a March 2026 supply-chain attack involving the LiteLLM package, potentially exposing contractor PII — review their post-breach disclosures before onboarding. → Apply: mercor.com

micro1 — Crossed $100M ARR in December 2025, up from $7M at the start of that year. $35M Series A at a $500M valuation in September 2025, led by 01 Advisors. Selective AI-powered vetting through "Zara" certifies approved professionals. Trainer/annotator $20–$40/hr, evaluator $20–$65/hr, engineering $50–$150/hr. Direct competitor to Mercor and Scale. → Apply: talent.micro1.ai

Meridial (by Invisible Technologies) — Invisible reached $134M revenue in 2024 and raised $100M in September 2025. Meridial requires a degree and specialized knowledge in law, STEM, finance, etc., with 400+ projects open at last count. Hires from US, Canada, UK, Ireland, NZ, Australia. Strong fit if your domain is specialized rather than generalist. → Apply: meridial.com

One platform is a job. Two is a hedge. Three is a portfolio.

The June cohort is filling up quickly.

If this week's piece resonated — if you read the "build, don't license" section and felt it land — the cohort is where the build actually happens. Twelve weeks. Fifteen seats. Three deliverables:

This is the cohort for the operator who has decided that "I've tried a lot of tools" is not a strategy and "I built one that works" is.

→ Apply: portlev.com/cohort

Open a fresh chat with whatever LLM you use most. Paste in something you wrote last year — a memo, a strategy doc, an email that got a real response.

Ask the model to tell you, specifically, what it would have edited out.

Then look at the list and ask yourself which of those edits would have made the document worse — not by your standards, but by the standards of whoever you wrote it for.

That list is the map of your variance.

That is the part of you the model wants to smooth away.

Keep the map.

Reply to this email. I read every one.

Yuri Kruman Founder, Portfolio Leverage Co. · 3x CHRO · AI Trainer: OpenAI, Meta, Microsoft PortLev.com · LinkedIn · AIWageGap.com

"The canary did not die because the cage was small. The canary died because the air it was breathing was being made slightly worse, continuously, in a way the miners could not feel until it was too late. The miners did not need a better canary. They needed a different relationship with the air."

— After Ludmila Praslova, The Canary Code

© 2026 Portfolio Leverage Company. You're receiving this because you subscribed to The Leverage Brief. Unsubscribe in one click.