The name had to wait for its moment

I read William Gibson’s Idoru in San Francisco in 1999. In the novel, eigenheads are virtual personality constructs – identity models derived from mathematical analysis of human behavior. Gibson took a term from linear algebra and used it for something nobody had a word for yet. The concept stuck, and I registered eigenheads.com that same year. What for, I didn’t know. Only that it meant something.

The math underneath

The scientific side came later. Eigenheads descend from eigenfaces, a face recognition method Turk and Pentland introduced in 1991. It reduces a face to its essential features, the eigenvectors, and in that reduced space identity becomes computable.

The same prefix runs through all of modern AI. Linear algebra is the bedrock of every neural network and every large language model. When you talk to an AI, you’re watching linear algebra at massive scale. Gibson intuited in 1996 what the math proved decades later: identity, pattern and meaning can be computed. Not perfectly, not completely. But enough to change everything.

Why this feels familiar

I had internet access and an email address at university in 1993. Back then that was as esoteric as AI was ten years ago. By the late nineties the internet was in every newspaper, and many executives were convinced it would pass and business would go on as usual. The dotcom crash seemed to prove them right. Then startups and competitors who had taken the technology seriously started eating into their business. The pace never slowed down. It sped up.

AI follows the same pattern, only faster. The winners won’t be the companies that wait for the perfect use case. They will be the ones that try now, learn what works, and understand enough to recognize the revolutionary idea when it shows up.

From a name to a firm

I started eigenheads as a consulting practice in 2014. The name finally had a purpose: analytical precision – eigen – and the people who make it useful – heads.

Then AI accelerated. Today we work with models that hold conversations, reason, generate and sometimes surprise the people who built them. Gibson mapped this territory in fiction, and now we live in it.

What we’ve learned

The more I work with AI, the more I recognize Gibson’s world. Not the dystopia, but the texture: human expertise and machine capability blurring at the edges.

We’ve run discovery workshops with reinsurance consultants and audit firms. We’ve built production applications for industries most AI companies have never touched, and we’ve developed open standards for material identity data. Every engagement shows us a bit more clearly where the edge really is.

One pattern holds every time: the people who know their domain deeply get the most out of these systems. Specialized knowledge doesn’t lose value when you connect it to a model. It gains value, because suddenly it scales. Our job is to make that connection: fast, live, without theatre.

Why we do this

We don’t do demos or proofs of concept that end up in a drawer. We do actual work: exploring, building, and putting things into operation that change how an organization works.

Nobody knows what comes after today’s models, and I don’t either. But the question will stay the same: how do people who know something work with machines that can compute it? We’ve been working on that question since 1999. The math has only just caught up.