The Compressing Generalist Layerwhy the ends defend what the middle cannot

At Staff on the IC track, depth starts to diverge. Two distinct archetypes emerge, each irreplaceable in a different way.

The design spectrum stretches from abstract to concrete. At the upstream end, the work is pre-artefact: problem framing, research strategy, behavioural reading, brief interrogation, the translation between user truth and business consequence. At the downstream end, the work is post-framing: motion calibration, code-integrated interaction, design systems at the token and governance level, production-level craft. Between them sits the generalist layer: standard visual production, layout, common interaction patterns, typical component assembly, the work most practitioners do most of the time, and the work most of the discipline was trained to do well.

Across the last several years, accelerating through 2025, the generalist layer has become the part of the spectrum AI tooling compresses fastest. Generative tools turn out passable visual mocks, typical interaction patterns, and standard layout systems with a speed and variety a human generalist cannot match. That layer still needs judgement, but it is no longer the source of depth-based career protection: a senior designer specialising in above-average mid-spectrum work now competes with tooling that produces acceptable mid-spectrum work at a fraction of the cost.

The two ends are harder to compress because they rely on what AI tools do least well. Upstream work requires contextual judgement: reading what a brief actually is beneath what it says, choosing which problem a team should solve when it could solve many, and holding the organisational trust that lets strategic recommendations be accepted. These do not generalise across projects or organisations the way generalist craft does. They depend on accumulated history, earned relationships, and pattern recognition built over years.