Anthony Guzzardo

I'm a software engineer who hasn't written a line of code by hand in four months. I work with AI all day. I can feel it changing how I think and honestly that's part of what got me here, but not in the way you'd expect.

This started as a science forum. I wanted to build a place where researchers could collaborate on new theories and maybe get funding. A Reddit for STEM with actual money attached. Physics and math primarily, which is funny because I'm genuinely bad at math. It's one of my insecurities. I'll never be good enough to do the work myself but I like to follow it and that's enough to keep me near it. So I built a site that pulls from arXiv and NASA every night and ranks what's trending. Black holes, quantum gravity, that sort of thing. A cron job runs and I let it ride.

Then I got a different idea. What if you could build a system where someone journals with deep questions every day and the system eventually starts generating its own theories? Like an Einstein machine. You write and write, it listens, it doesn't bias you, and at some point something clicks and it produces a genuine breakthrough. I built that. Added predictions, suppressed questions, behavioral signals. The whole thing.

It was broken. Circular loop that graded itself. LLMs can reason but they can't generate what doesn't exist yet. I was proud of the architecture and the architecture was fundamentally flawed. That was a rough few days.

But in tearing it apart I found something I wasn't looking for. If you strip away the genius-machine ambition what you're left with is a system that measures how someone thinks through how they type. Not what they say. How the keystrokes land. The pauses, the deletions, the rhythm, the revision patterns. And if you do that every day for long enough against the person's own baseline you have something that doesn't exist anywhere in the research literature. A longitudinal cognitive instrument embedded in a daily practice that someone actually wants to do.

I started reading the research and found out why nobody had built this. The people who study keystroke dynamics for cognitive assessment are measuring elderly populations who learned to type in their 40s. They're hen-pecking. The motor noise drowns out the cognitive signal. But there's a generation coming into the screening window who've been typing fluently since childhood. That's the population this instrument is designed for. And that window is closing because every input modality is being contaminated by AI assistance. Autocomplete, predictive text, copilots. The cognitive process being measured is being replaced by a different process that looks identical on the surface.

So now it's a race. Build the instrument, start collecting clean data, accumulate baselines before the unmediated input disappears. I'm testing on myself because someone has to go first. I built the entire pipeline. Rust for the signal math, Postgres for the storage, six families of signals computed every session. I'm the engineer and the only participant which is a real confound and I'll say that plainly. The instrument needs people who didn't build it. That's the next step and it requires collaborators I don't have yet.

I've always been a good typist so at least the modality fits. And honestly at the very least this whole thing got me writing again. I built it because it's what I love to do and because I think it should exist.