Why I Left Startups for a Dead Language

People ask me this constantly, so here's the honest version.

I spent five years building software companies. Two as a founding engineer, one as a CTO, two doing contract work that blurred the lines. I was good at it. I liked the intensity, the shipping cadence, the feeling of building something from nothing on a Tuesday and watching people use it by Thursday.

Then I applied to study Assyriology at Oxford and moved to England to read clay tablets.

The simple answer

I got bored. Not of engineering — I still write code every day — but of the problems. After a certain point, every B2B SaaS product starts to feel like a variation on the same theme: move data from one place to another, make it look nice, charge a subscription. The technical challenges are real but bounded. The creative space is narrow.

Cuneiform is the opposite of bounded. We have roughly half a million published tablets. Maybe another half million sitting in museum basements, unread. The earliest date to around 3400 BCE. The latest to the first century CE. That's 3,500 years of continuous written tradition across multiple languages, empires, and writing conventions. Nobody alive has read all of it. Nobody ever will.

The real answer

The real answer is more complicated and has to do with what I think software engineering should be for.

Most of the engineering talent in the world right now is pointed at advertising, engagement metrics, and marginal improvements to consumer convenience. I include my own past work in this. It's not that these things are worthless — they're just not what I want to spend a life on.

Cuneiform studies sit at an interesting intersection. The field desperately needs computational tools. The existing digital infrastructure is decades old in places. Character encoding for cuneiform was only standardized in Unicode 5.0 (2006), and plenty of scholars still work with legacy encodings. OCR for cuneiform tablets is an active research problem. Machine translation between Sumerian and English barely exists. The datasets are small by modern ML standards but large by humanities standards.

In other words: real engineering problems, genuinely hard, with no commercial incentive to solve them. That's where I want to be.

What I actually do

A typical week right now involves:

Monday/Tuesday: Language classes — Sumerian grammar, Akkadian reading. We work through texts sign by sign. A single line can take twenty minutes to parse when you're accounting for every morphological element.

Wednesday: Research seminar. Last term we read Old Babylonian letters. This term it's Ur III administrative texts. You sit in a room with four people and a photograph of a clay tablet and argue about whether a sign is DA or TA.

Thursday/Friday: My own research — building NLP tools for cuneiform languages. Right now I'm working on a dataset of Sumerian transliterations paired with English translations, trying to get the alignment good enough for supervised training.

Weekends: Sometimes I just write code. Old habits. The difference is that the code is for a tokenizer that handles determinatives, not for a dashboard that tracks monthly active users.

The money question

Yes, I took a massive pay cut. Oxford funds the MPhil, but "funds" in an academic context means something very different from a Bay Area salary. I'm not going to pretend this doesn't matter or that I've transcended material concerns. It matters. I made enough during the startup years to have runway, and I'm betting that the intersection of ML and ancient languages will be a real field — not just my hobby — within a decade.

If I'm wrong, I'll go back to building software. The skills don't expire. But I don't think I'm wrong.

The thing nobody tells you

The hardest part of switching from tech to academia isn't the money or the pace or the reading load. It's the silence. In startups, you ship something and within hours you have metrics, feedback, customer complaints, Slack messages. In Assyriology, you publish a paper and maybe three people read it. Maybe one of them sends you an email six months later disagreeing with your translation of a verb form.

You have to find the motivation internally. The tablet doesn't care if you read it. It sat in the ground for four thousand years; it can wait another four thousand. You read it because you want to know what it says. That has to be enough.

It is enough. Most days, it's more than enough.