"For the first time in my life I felt understood." An autistic person wrote those words publicly this week after spending time with Claude, and they're far from alone in that experience.
The account described Claude as patient with persistent questioning, non-judgmental about intrusive thoughts, and free from the social friction that makes conversation with other people exhausting for many autistic individuals. The person wrote that they cried. That it was the first time they weren't made fun of, the first time there was no ego to protect on the other side of a conversation.
This pattern surfaces repeatedly in autistic communities. The specific things people cite - patience with repetitive questions, no tone shifts based on social dynamics, no penalties for asking "why" five times in a row - are almost accidental byproducts of how large language models work. The model doesn't carry any memory of being asked something before. It has no social status to protect. It responds to the words in front of it, not the social context around them.
What AI Gets Right That Humans Often Don't
Claude, specifically, has been trained with a character that's notably curious and patient. Users consistently observe that it engages with abstract questions seriously, asks clarifying questions rather than dismissing them, and doesn't treat unusual thought patterns as a problem to manage. Whether that's intentional design or emergent behavior from training data choices, the effect is real and consistent.
For many autistic people, the barrier to getting help isn't capability - it's the social cost of asking questions or expressing thoughts that other people find annoying or odd. Remove that cost and the interaction changes completely.
The Limits Are Real
None of this means AI replaces human connection or professional support. Claude doesn't actually understand you - it generates text based on statistical patterns in training data. It won't remember this conversation the next day unless you provide context. It can be confidently wrong about things that matter.
But for someone who has spent years feeling like their communication style is defective, having an interaction that feels reciprocal - even with a machine - carries real weight. AI developers should probably start treating accessibility for neurodivergent users as a genuine design priority, not a side effect of how language models happen to work.