A recurring observation from the past six months of using generative tools in production work: AI fails at precisely the moments where a designer would say “it just felt wrong.”
This is philosophically interesting. It means our highest-value judgment is pre-linguistic — it operates below the level of articulable criteria.
What the signal points to
If you can fully specify the criteria for good design output, an AI can approximate it. The gap is always in the unspecified criteria — the ones that make the work feel authored rather than produced.
This suggests the most defensible position for a design practice is not technical sophistication but a cultivated sensitivity that can’t be fully documented.
The paradox: the less you can explain it, the harder it is to replace.
Calibration
The studios that will feel this least are the ones already operating at the level of conceptual authorship — where the design is the least important part of what they deliver, and the thinking is the actual product.