The sanctuary problem
Traditional programming operated on a simple contract: write explicit instructions, get predictable results. For neurodivergent developers—15-20% of the workforce, higher in Gen Z—this wasn't just convenient. It was the only professional environment that didn't require constant Theory of Mind calculations.
A compiler has no hidden agenda. return 0; means exactly one thing. Human interaction doesn't work that way. "I'm fine" could mean happy, angry, tired, or leave me alone. Programming eliminated that ambiguity entirely.
Now AI coding tools are reintroducing it.
What's actually changing
GitHub Copilot, Amazon CodeWhisperer, and similar tools don't execute instructions—they interpret intent. They operate on context, implications, and probabilistic pattern matching. They hallucinate. They give verbose explanations when you need a boolean. They require prompt "massaging."
The source code is now subject to linguistic ambiguity. You're essentially negotiating with a statistical model to write a SQL query.
Some neurodivergent developers report sensory overload from constant inline suggestions. Others find the tools disable their executive function by interrupting flow states. The very features designed to help—autocomplete, context awareness, predictive text—can undermine the cognitive patterns that made these developers effective in the first place.
The enterprise angle
This matters beyond individual productivity. Companies like Microsoft and EY report that neurodivergent teams generate 60-80% more innovation ideas when AI tools reduce cognitive load. But that only works when the tools align with neurodivergent working styles.
The alternative data: Neurodivergent developers are 30% more productive and less bias-prone in traditional environments. They excel at edge-case detection, pattern recognition, and bias mitigation—exactly what AI safety requires. Exclude them from AI development, and you risk regulatory failures, discriminatory outcomes in hiring algorithms, and blind spots in model audits.
What to watch
The industry is moving toward "English as the programming language." Fine for neurotypical developers. Potentially catastrophic for those who chose programming precisely because it wasn't English.
Worth noting: UiPath launched a Neurodiversity People Unity Council. Some vendors are building accessibility features. But most AI coding tools assume neurotypical cognitive patterns.
The real question: Can we build AI assistants that preserve determinism while adding capability? Or are we optimizing for the majority and calling it progress?
History suggests the latter. We'll see if this time is different.