Over twenty years ago, Scandinavian silent movie confronted me with a semiotic problem I couldn't let go of: where does meaning break down between image and language, and what fills the gap? Silent movies for example forced directors to communicate without language, and revealed exactly where the image fails: inner states, abstraction, time.
That was an aesthetic problem sitting at the intersection of semiotics, linguistics and psychology. With multimodal AI models, I think it is still a question worth to be answered. These models are trained on billions of image-text pairs, not because images and text say the same thing, but because meaning lives in how they meet.
I started as a language editor for technical documentation in Swedish, working at the point where complex knowledge has to become usable language. That question of transfer: how does expertise become accessible? It never left me.
From there: designing sales training for automotive dealerships navigating the shift from B2C to B2B. Then contract logistics, where new employees, often temporary workers, had to become competent fast, under real operational pressure, at go-live. Then People & Culture Lead at an SAP consultancy, with strategic responsibility for repositioning how an organization builds capability. Today, I work in customer enablement at a software company, supporting software adoption at the intersection of technology, process change, and organizational development.
The setting changes. The question doesn't: what do people need to be able to do when the context shifts, and how does that capability actually come about?
AI sharpens that question considerably. If AI handles the content, the process of making sense of it becomes the real competency. But that also means understanding the technology itself, not at an engineering level, but well enough to use it responsibly. Knowing why a model hallucinates, or answers with confidence while being factually wrong, changes how you work with it. And every imprecise query, every unnecessary iteration consumes compute, energy, and water. Understanding AI well enough to use it well is not just a professional competency. It is a form of responsibility.
Most people working on AI ask about capability: what can the model do? I'm more interested in the other side: what does it require of us? What competencies become more important, which ones shift, and how do organizations build them deliberately, and not by accident?