We talk a lot about the democratisation of manufacturing. But spend time in this community, really spend time, and a harder truth becomes visible.The machines have never been better. Bambu Lab's A1 is practically a household appliance at this point. Self-calibrating, smartphone-connected, quiet, fast. The hardware barrier has been dismantled.And yet printers go idle. Filament spools go stale. Machines end up listed at half their purchase price.Because the barrier was never the machine. It was always the model. A 3D printer without a 3D model is an expensive paperweight. And creating a functional model, one that is dimensionally correct, structurally sound, ergonomically sensible, and actually printable, requires skills most people don't have and have no particular reason to acquire. The learning curve for CAD isn't steep. For most people, it's vertical.So users do what's rational: they come here, to MakerWorld. They look for something that already exists. When it doesn't exist in exactly the form they need, which is often, they hit the wall. The platforms have a partial answer: some parametric models with exposed variables. Real engineering under the hood, a more or less friendly interface on top. MakerWorld's Parametric Model Maker (with Fusion support) is a genuine step forward. I've been experimenting with it; the Gridfinity rack I published is an example, and the potential is real.But the limitation is structural. The user can only customise what the creator already anticipated. The form, the logic, the constraints, all decided in advance by someone who knows what they're doing. The user is a passenger, not a driver. And the community of creators who build these models is smaller than it looks. Attrition is brutal. Someone discovers they can model, publishes a few things, gets little visible traction, and quietly disappears. Those of us who've stayed are genuinely exceptional — and I mean that in the literal sense. We are the exception that the system depends on but cannot reliably reproduce. AI generative tools are the obvious next candidate. Text-to-3D, image-to-3D, the forms they produce can be remarkable. For decorative objects, sometimes genuinely surprising.But here's the problem: current AI is fundamentally stochastic. It generates geometry from statistical patterns. It doesn't know what a tolerance is. It doesn't know that the wall thickness matters, or that the gap between two mating surfaces is measured in tenths of a millimetre and determines whether the thing works or binds or falls apart.You cannot derive physical correctness from pattern-matching alone. The two are different in kind. So where does that leave us?Form generation: AI has it. Deterministic engineering logic: CAD has it. Distribution: platforms have it. Manufacturing knowledge: simulation tools have it. Conversational interface: large language models have it.What no one has yet is the architecture that connects them — the system that decides when to be generative and when to be deterministic, when to ask the user a question and when to apply a rule silently.I've published a longer piece exploring this in detail — the full argument, the map of tools, the research gap, and what I think convergence between the major players might look like. I'd genuinely love to hear what this community thinks. You've all lived this from the inside. What am I missing? → [Full article: The Machine That Waits] Jorge Rui Silva — Mechanical Engineer, Industrial Designer