What "neurodiversity at work" programs get wrong: assimilation sold as inclusion

When SAP launched Autism at Work in 2013, it created a template that every enterprise neurodiversity program since has copied, refined, and branded. Microsoft’s Autism Hiring Program, JPMorgan Chase’s Autism at Work, Dell’s, EY’s, the string of programs that followed — each one announces its hires, generates favorable press, and celebrates the narrative that autistic people are untapped technical talent waiting for a company brave enough to see them. The press releases are polished. The programs are often genuinely well-intentioned. And they are, almost without exception, selling assimilation and calling it inclusion.

I write this as someone who builds and manages technical teams and as a neurodiversity-affirming parent of an autistic child. I have not personally gone through one of these hiring tracks. That’s a vantage limit I’m honest about: the autistic people who lived these programs are the authorities on what they feel like from the inside, and the marginalized autistic critique of these programs is the voice that tech and business press consistently fails to amplify. What I can do is name the structural pattern I see as a manager, connect it to the autistic research literature that explains the harm, and argue that the programs are not worthless — they are incomplete and mislabeled in a way that matters.

The PR pattern

The corporate neurodiversity program follows a predictable arc. A company announces a hiring initiative, often with a partner organization that handles sourcing and assessment. The announcement emphasizes the business case: autistic people have pattern-recognition skills, attention to detail, sustained focus on repetitive tasks — the savant-adjacent framing that selects for a narrow slice of autistic people who happen to match a technical role description and filters out the majority who don’t. The press cycle celebrates the program. A few autistic employees are profiled. The company receives diversity accolades. And then, quietly, one of two things happens: the program stays a pilot indefinitely, never scaling beyond its initial cohort, or the autistic hires are tracked into specific role types — QA, data annotation, repetitive testing — that segregate them from the general engineering pipeline.

The savant framing is not incidental. It is the entry condition. The program only recruits autistic people whose cognitive profile maps to a pre-existing technical need, which means the program isn’t valuing autistic cognition as such — it’s valuing a specific marketable skill that happens to correlate with some autistic people. The autistic person who is brilliant at policy analysis, narrative structure, or systems ethics but not at sustained repetitive testing doesn’t fit the program. The program has not decided autistic people are valuable. It has decided that a narrow band of autistic people are valuable for a narrow band of work, and it has called that inclusion.

The press release is the product. The program is often the pilot. And the success metric — when it is stated at all — is retention: did the autistic employee stay? Retention measured without examining what the employee had to do to stay is a metric that rewards masking. The autistic person who learned to soften their communication, attend the social events, suppress their sensory discomfort, and perform neurotypical workplace norms is counted as a success story. The autistic person who burned out and left is counted as a statistic the program prefers not to publish.

Assimilation as the operating model

Strip the branding away and the implicit success criterion of these programs is clear: did the autistic employee adapt to us? The onboarding materials, the buddy systems, the social coaching — these are designed to help the autistic employee learn the existing workplace culture. The workplace itself is treated as fixed. The autistic employee is the variable to be adjusted.

This is masking labor. The autistic burnout literature — Raymaker and the AASPIRE collaboration, the work that has built the most rigorous account of what autistic burnout is and what causes it — identifies chronic masking as a primary mechanism of harm. Masking is the sustained performance of neurotypical social norms by someone whose natural social register is different. It is cognitively expensive, it accumulates, and it produces the exhaustion, withdrawal, and functional loss that autistic people describe as burnout. I wrote about this mechanism in detail in the context of knowledge work. A workplace program that requires masking as the condition of participation is prescribing the harm it claims to address. The program that offers an autistic employee a job on the condition that they learn to communicate differently, socialize on demand, and tolerate a sensory environment designed for neurotypical defaults has not included that person. It has contracted them into a performance and called the performance belonging.

The double-empathy problem makes this even clearer. Damian Milton’s framing — that the communication breakdown between autistic and non-autistic people is bidirectional, not a deficit located in the autistic person — has direct implications for workplace design. I’ve argued elsewhere that this reframe lands in the artifacts of software work, including pull requests. If the friction is bidirectional, then “fix the autistic person” is the wrong lever. The non-autistic team members also have a communication style that needs examining. The neurotypical preference for indirect, hedged, socially-softened feedback is not neutral — it is one register among many, and it is the one the autistic employee is being asked to learn. A program that locates the deficit entirely in the autistic employee and calls the one-sided adaptation “inclusion” has not done inclusion. It has done assimilation with better marketing.

Disclosure as the tell

The structural giveaway is disclosure. These programs typically require autistic employees to disclose their diagnosis as a condition of entering the specialized hiring track. The track itself is labeled — neurodiversity program, autism hiring initiative — which means the employee’s diagnostic status is legible to the organization in a way it is not for employees in the general pipeline. This produces two consequences. First, the employee enters a labeled pipeline, which research and autistic community accounts consistently describe as carrying thinner advancement opportunities and persistent “special needs” framing from managers and peers. Second, the employee cannot later un-disclose; the label follows them.

This is not a minor structural detail. Disclosure in software workplaces is a high-stakes calculation, and autistic employees weigh it against documented costs: being passed over for leadership, being assumed incompetent at social-collaborative work, being treated as a workplace accommodation case rather than a contributor. A program that requires disclosure as the entry condition has not solved that problem. It has made the disclosure mandatory and called it access.

Contrast this with the actual goal. A workplace where literal, direct communication is the default norm — where PR templates specify what feedback looks like, where async-first workflows mean nobody has to process spoken language in real time to participate, where meeting norms are explicit rather than intuited — is a workplace where an autistic employee doesn’t have to disclose to be accommodated, because the accommodation IS the default. The accommodation is universal design, not a specialized track. Inclusion that requires you to out yourself to access it is a tracked program. Inclusion that redesigns the default so the question of disclosure becomes irrelevant is the real thing.

What real inclusion would look like

Real inclusion redesigns the artifacts, not the people. The double-empathy reframe — which I wrote about in the context of code review friction — tells us that the breakdown isn’t located in the autistic person’s communication style. It’s located in the interaction between two different cognitive styles, both of which are doing something legible and reasonable, and neither of which has been given a protocol designed for the gap between them. So you design the protocol.

Pull request templates that separate what changed from why it changed and from what trade-offs were considered give autistic contributors a structure where their natural mode — explicit, literal, thorough — is the expected one. Async standups in writing instead of live verbal status rounds remove the real-time social decoding tax that makes the daily standup a daily micro-burnout. Review checklists that score the change — correctness, testability, architectural fit — and not the commenter’s tone, with an explicit team norm that says “don’t read into tone, we mean what we write and nothing more,” remove the reciprocal burden where autistic bluntness gets flagged as aggression and neurotypical indirectness gets flagged as evasiveness. None of this is charity. It’s better protocol design. It helps everyone, and it specifically stops punishing the people whose cognitive defaults were never accommodated by the original protocol.

Sensory environment defaults matter. Open-plan offices with fluorescent lighting, ambient conversation, and no escape routes are not a neutral baseline that autistic employees are asked to tolerate as a personal limitation. They are a hostile baseline. Quiet zones, remote-first defaults, headphones-as-acceptable, lighting that doesn’t strobe at the edge of perception — these are not accommodations extracted through a disclosed medical process. They are better defaults. The fact that they require a disclosed medical process to access is itself the problem.

Evaluate work, not “culture fit.” Culture fit is a euphemism that, in practice, means “do I feel comfortable around this person,” and “comfortable” is doing a lot of unexamined work there. It is the mechanism by which homogeneity reproduces itself while sounding like a virtue. Evaluate the code. Evaluate the design decisions. Evaluate the documentation, the research, the ability to find edge cases that others missed. These are evaluable. “Culture fit” is a vibe metric, and vibe metrics optimize for the dominant group’s comfort.

Manager training on the double-empathy frame: if your managers still believe the autistic person in the room is the one with the communication deficit, you have not trained your managers. Both parties have a communication style. The friction is in the gap, and the gap is a shared responsibility. Managers who understand this can de-escalate review conflicts by naming the dynamic instead of defaulting to “the autistic person needs to soften their tone.” The tone was probably fine. The recipient’s interpretation framework was the issue, and that’s a two-way street.

Promotion paths must be identical to the general track. Not a parallel track labeled “neurodiverse” with thinner scope, capped seniority, and the permanent asterisk of “hired through the program.” A labeled pipeline is a segregated pipeline. If your neurodiversity hiring program funnels people into QA, data entry, “focused technical contributor” roles that never lead to staff or principal or architect or management, you have not built an inclusion program. You have built a sorting mechanism that finds autistic talent and routes it away from seniority. Real inclusion means the autistic person hired through the program is on the same ladder, scored by the same rubric, eligible for the same scope, and competing in the same promotion calibrations as everyone else.

The success metric flip

Here’s the test. Measure success by whether autistic employees reach the same seniority and scope as their peers over five years. Not by whether they “adapted well.” Not by whether their managers report they’ve “improved their communication.” Not by retention numbers that don’t distinguish between people who stayed because the workplace worked for them and people who stayed because they had no other option and learned to mask harder. A program measured on masking-adaptation will optimize for masking. It will reward the autistic people who learned to pass, and it will lose the ones who couldn’t or wouldn’t, and it will call that success. A program measured on seniority parity will optimize for the workplace changes that produce parity. It will force the question: why are autistic employees not reaching staff engineer? Why are they not becoming tech leads? Why are they not managing teams? And the answer will be structural, and the fix will be structural, and the program will either adapt or reveal itself as theater.

Honest about the tension

I want to be clear about something, and I want to be clear about it from a position that acknowledges its limits. Some autistic people credit these programs with giving them their first foothold in the industry. I’ve heard this. I believe it. The programs are not worthless. They are incomplete and mis-labeled. The difference matters. The critique here is not “shut them down.” It is: stop calling assimilation inclusion, and finish the job. Finish the job by redesigning the workplace around autistic cognition instead of redesigning the autistic employee around the workplace. The foothold is real and it shouldn’t require the rest of the cost.

And I have to be honest again about where I’m standing. I’m a manager of technical teams and a parent of an autistic kid. I’ve designed hiring processes and I’ve redesigned team norms. I have not personally been through one of these hiring tracks. The autistic people who lived these programs from the inside — who went through the intake assessments and the “social skills” modules and the labeled pipelines — are the authorities on what that experience actually felt like and what it actually cost. This essay names the structural pattern from the manager’s seat, and it points at the literature on double empathy, autistic burnout, and workplace disclosure, but it is not and cannot be the inside account. Those exist. Read them.

Coda

A neurodiversity program that changes the autistic employee and not the workplace has the relationship backwards. It is accommodation theater performed on the employee’s body and cognition while the built environment — the protocols, the sensory defaults, the promotion structures, the evaluation rubrics — stays exactly as it was. Inclusion is when the default works for autistic cognition without anyone having to disclose, mask, or enter a labeled track to access it. If your program requires disclosure as the entry condition, it is not inclusion. It is a separate door to a smaller room, and the person who walks through it is asked to be grateful the door exists at all. The door is not the point. The building is the point. Redesign the building.