The distinction sounds subtle. It is not.

Generative design is a search process. You define constraints — span, load, material budget — and the algorithm explores the solution space and returns candidates. The engineer selects. Judgment happens at the output end.

Predictive design inverts this. The model doesn’t return options. It returns a forecast: given these site conditions, this programme, this structural system, here is the expected performance across every metric simultaneously. Embodied carbon. Seismic response. Lifecycle cost. Deflection under service loads. Before you have drawn a single line.

The engineer’s role shifts from selector to envelope-definer. You are no longer choosing between what the machine found. You are specifying the boundaries within which the machine operates — and then interrogating its conclusions.

This changes what engineering judgment means. The skill is no longer in reading the output. It is in knowing which questions to ask before the model runs, and which answers to distrust when it returns.

PE stamps were designed for a world where the engineer performed the calculation. When the calculation is performed by a model trained on ten million prior structures, the stamp certifies something different. Nobody has agreed yet on what.

That gap — between where the tools are going and where the profession’s legal and ethical frameworks currently sit — is the territory this platform tracks.

The singularity for structural engineering is not the moment AI designs a building. It is the moment the engineer can no longer fully explain how the answer was reached. That moment is closer than most practitioners realise.