For about three years, Polina and I had an almost nightly habit of talking about whatever she was studying — psychology, philosophy, memory, learning, behaviour, identity, consciousness, emergence. Why people do what they do. Why behaviour changes with context. Why memory is reconstructive. How something as apparently unified as a person emerges from an enormous number of interacting processes.
Then Polina started working in behaviour support. And another problem showed up.
Years earlier I'd studied machine learning. I pulled the laptop back out and started trying to build a statistical prediction engine for behaviours of concern. It was wildly ambitious — which made it extremely useful, because human behaviour destroys naive models.
The same action can mean different things. Different internal states can produce the same visible behaviour. History matters. Environment matters. Relationships, sequence, timing — all of it matters. Observation changes interpretation. The boundary you draw around the system matters. And uncertainty is everywhere.
I kept hitting the same wall. Before I could ask what will happen next?, I had to answer something more fundamental: what exactly am I representing?
I'd seen it before. In the conveyor network. In shortest paths. In the behavioural model. In neural networks. Different problems, different domains, different terminology — and underneath all of them the same shape: things in states, capable of relationships, changing through interaction, forming recurring patterns, and sometimes producing patterns stable enough that we start treating the pattern itself as a thing.
That recognition was the beginning of Primitive. Not the company — the idea.
Because the work keeps returning to one question: what's the smallest thing we need to assume? Not the easiest thing to name. Not the first thing an observer sees. The primitive — the irreducible element you can build the useful description from.
Primitive doesn't mean unsophisticated. It means foundational.
Primitive runs across three overlapping modes. Research — investigating representation, information, relationship and emergence. Engineering — turning those ideas into computational systems. Applied work — using the results on real operational problems. Each keeps the others honest: theory without implementation drifts, engineering without theory becomes accidental complexity, and commercial work without either becomes another wrapper. Primitive lives in the overlap.
The same systems thinking runs through everything we build, in three directions.
Primitive studies how complex systems work. WHATBIT fixes the part that matters. Elsewhere builds new systems in the physical world.
Down, in, out. One method, three moves.
Find the thing beneath the thing. Start there.