Every era of enterprise software produced a profession. Mainframes produced the systems analyst. ERP produced the process consultant. The cloud produced the platform engineer. Each profession existed because the previous generation of tools had reached the limits of what untrained people could safely operate.
We are at that point again. The systems we are now asked to build — semantic layers, agentic workflows, decision architectures, trust-grade automation — do not have a profession yet. There is no university degree for them. There is no certification worth its paper. There is a small, scattered group of practitioners who learned by doing.
What the new craft actually is
Four disciplines are emerging, and they do not map cleanly onto any existing job title.
Decision Designers treat the decision itself as the unit of design. They map the signals that should trigger it, the context that should inform it, the people and agents that should make it, and the action that should follow. Part product designer, part operations researcher, part organisational psychologist — and none of those alone.
Semantic Architects build the layer where meaning lives. They decide what customer, order, risk, active mean in this organisation, in this quarter, under these constraints — and encode those decisions in a form every system above can reason from. Their work is slow, political and structurally underrated. It is also the highest-leverage role in a modern enterprise.
AI System Engineers wire reasoning into the operational fabric. They are not prompt engineers and they are not MLOps. They design how a model is grounded, how it is governed, how its outputs become actions, and how it fails safely when it does fail.
Trust Engineers build the properties that let people believe what the system says — provenance, reversibility, observability, accountability. They are the reason a board will eventually let an automated decision stand without a human stamp on top of it.
Why this cannot be self-taught at scale
Every craft has a stage where one or two practitioners can carry it alone. That stage is over for this one. The organisations asking for these systems are too large, the stakes too high, and the gap between "I built one for my team" and "I can build one for a 40 000-person enterprise" is not bridged by a weekend course.
Real craft is transmitted in three things: a shared vocabulary, a body of patterns, and a community of people who have already made the mistakes you are about to make.
None of those exist by accident. They have to be designed, written down and taught — over and over, by people senior enough to know where the patterns break.
What we are building
Not a bootcamp. Not a certification mill. Not a scrollable content library. The smallest, sharpest community of practice we can — and writing down the patterns slowly enough that they survive being taught. The systems are ready. The tools are ready. The craft is the missing piece.