Thanks for sharing my thread!

I'm posting a series this month on multiple social media, so I have other recent threads on autism.

You can also find the full series on dev.to/raphink/impr... (will be updated as I keep posting until the end of this month).

IT didn't create more autistic people. It created conditions where autistic people could function, contribute, and sometimes thrive — without anyone necessarily noticing why. The diagnosis rates are rising because awareness is rising, not because something new is happening.

Pattern recognition, deep focus on narrow problems, a preference for systems that behave predictably, a low tolerance for ambiguity that, in code, is actually a feature.

The cognitive traits that make social navigation harder — needing explicit structure, struggling with unspoken rules, finding small talk costly — are often the same traits that make technical work easier.

Autistic people are more likely to have autistic children, and certain fields, IT among them, have been quietly concentrating people with this neurological profile for decades. Long before most of them had a name for it.

When the psychiatrist finally explained why our son was on the spectrum, I remember thinking: he's describing me.

That's not an unusual story. Autism has a well-documented genetic component.

My wife worked for years as a psychologist with autistic children. Shortly after our second child was born, she started noticing patterns in him that she recognized professionally. It took ten years to get a formal diagnosis.

That wasn't a trend. That was a room full of people who had been waiting for the door to open.

So: is there more autism in IT? Or does IT just make it easier to be autistic?
Probably both, and they're not contradictory.

Three years ago, I created a channel called hashtag #autism on my company's Slack. I wasn't sure what to expect. On the first day, about 20 people joined, roughly one in six of the company at the time. Four personal testimonies came in within hours. Dozens more wanted to understand better.

🧵

When something new appears, it rarely arrives in isolation: the system has already indexed something adjacent, something that rhymes, something from three fields over. Pattern recognition and unexpected correlations come almost automatically, not as a skill but as a byproduct of the constant scan.

Talking about it at length isn't ignoring the other person, it's the only moment the scanning stops.

There is an upside to all this collecting though. A brain that has spent decades gathering fragments from unrelated domains ends up with an unusually dense web of cross-references.

Special interests are the inverse of this. A domain I know deeply is a domain with no trapdoors. The intensity and relief that come with a special interest aren't enthusiasm, they're the feeling of ground that holds.

This is what produces the constant background hum: not anxiety about anything specific, but the nervous system permanently scanning for gaps in the map, because any gap is a potential trapdoor.

Not out of curiosity, but because there might be a joke about that actor later. Everyone will laugh. I won't follow. And unlike most social awkwardness, I can't fake it convincingly without the underlying data. The gap isn't uncomfortable, it's a predicted failure.

For me, the information-gathering isn't occasional, it's continuous, and it isn't very selective. Any piece of information feels potentially vital. If someone mentions an actor I don't recognize, I need to know who that is.

The result is a system that needs more explicit data to feel oriented. So it collects data, constantly.

Everyone experiences some anxiety about the unknown. The difference in autism is in the mechanism and the scale.

There's a well-documented phenomenon in autism research called intolerance of uncertainty (IU): the nervous system's difficulty tolerating absent information. The research links it directly to how the autistic brain builds predictions: less automatically, less reliably than in neurotypical brains.

My mother often reminds me that when I was a child visiting friends, I would tour every room and every corner of their apartment before I could sit down and play. I had no idea I was doing it. I just couldn't settle until I had a complete map.
That pattern never went away

The constant background hum

This is the fourth post in a series for autism awareness month. Previous posts covered the neurological vs. psychological distinction and what "spectrum" actually means (see link in the comments). This week: what runs in the background, all the time.

🧵

Thanks for the feedback! Hopefully this approach can be useful to more researchers.

I can even query from my phone since the memory backend is accessible from anywhere.

Claude builds links between databases (documents, persons) and I can also fix and improve the data in Notion manually whenever I need.

Notion becomes the memory layer: structured, relational, queryable. In the next session, Claude can read back everything accumulated so far. I can ask "what do we know about Louis's business dealings?" and get a brief drawn from months of work. I can ask "what's still unresolved?" and get a list.

Great benefit of using Notion: the results don't disappear when the session ends.

By the end of the session, I had a coherent picture of Louis's role in Edison's network — backed by primary sources, structured, and stored in a way I can actually build on.

Everything gets written to Notion: a structured page per document, with transcription, extracted data, confidence level, and links to related people. A running research notes page accumulates findings, contradictions, and open questions.

It checks each finding against the GEDCOM tree: is this person already known? Does this date contradict something we have? Is this a new individual to add?

For the interesting ones, it fetches the actual document images and reads them directly — handwriting included. It extracts names, dates, roles, monetary amounts, technical references, whatever is relevant.

Starting with a search for "Louis Rau" across the Edison Papers, Claude found hundreds of matching documents.

That's too many to analyse blindly, so the first pass uses abstracts only: Claude reads each abstract, scores relevance, and filters down to the documents worth examining closely.

👉 GEDCOM MCP — gives Claude live access to my family tree file, so it can cross-reference every name and date it finds against what I already know
👉 Filesystem MCP — for any local documents, scans, or notes I already have
👉 Notion MCP — writes structured results directly into Notion databases

I'm using Claude with a set of MCP servers — tools that let it interact with external systems directly during a session, rather than just processing text I paste in.

👉 Edison Papers MCP — searches the digitised archive, retrieves document abstracts, and fetches high-resolution scans for analysis