Generative massing is good at yield and bad at everything else

Generative site planning tools solve a real problem: they compress days of option testing into minutes and attach defensible numbers to each result. That is genuinely valuable at feasibility stage.

What they optimise

Unit yield, parking ratios, daylight compliance, and setback conformance. These are the constraints that can be expressed numerically, which is precisely why the tools handle them well.

What they cannot see

The quality of the street edge. Whether the entrance sequence makes sense. How the building meets its neighbour. These are the judgements that make a scheme good rather than compliant, and no current tool models them.

Using them honestly

Treat the output as a constraint envelope, not a design. The tool tells you what the site will bear; the architect still decides what should happen on it.