Ask a working architect what CAD is and you will hear “the software I draw in.” That answer was already wrong by 1963, and the gap between the answer and the reality is where the next decade of practice will be decided. CAD was never about drawing. It was always about relationships — and the question worth asking now is who, or what, gets to author them.

←TODAY: In 2026 every building, chip, and consumer object passes through a CAD model before it is fabricated; the model, not the drawing, is the deliverable.
→3012: The model becomes a queryable, versioned, partly self-authoring artefact — geometry that argues back.
Fulcrum: The insight only holds if you can see both ends — CAD began as a constraint engine, and it is returning to one, with the drawing as a side effect on both ends of the arc.

What it is: CAD — computer-aided design — is the system that holds the geometry of a thing to be made, together with the dimensions, tolerances, materials and relationships that let downstream tools act on it. In AEC it does three jobs at once, as PAZ’s own CAD concept panel puts it: it is the geometry kernel underneath BIM, the fabrication source for CNC milling and robotic printing, and the coordination surface across structural, MEP and façade trades. The crucial shift to internalise is that a contemporary CAD file is not a picture of a building. It is the source code of one. The drawings, the schedules, the BIM views — all of them are renderings of that single model.

Why it works: CAD holds because geometry became the lingua franca between disciplines that used to communicate through paper. The mechanism is mathematical and specific. Flat 2D drafting gave way to boundary-representation (B-rep) and constructive solid geometry (CSG), built on Euclidean geometry, parametric equations, B-splines and NURBS. In the 1990s feature-based and parametric modelling reframed the model as a tree of operations rather than a stack of lines — change one constraint, and the change propagates. That is the same idea Ivan Sutherland demonstrated in his 1963 thesis. The reason load-case simulation and topology optimisation are now routine rather than exotic is that the geometry stopped being mute: a column that knows it is a swept solid of a given material can be handed to CAE/FEA, to CAM/CNC, to PLM, without a human re-drawing it each time. From there the road forked into BIM, where geometry starts carrying semantic, lifecycle and cost data — which is precisely the thread the SIA 2051 norm and the BIM Label CH are trying to standardise for the DACH market.

Origins: The founding scene is small and bright: a CRT, a light pen, and Sutherland’s Sketchpad at MIT in 1963, which let a designer set parametric relationships between lines so that changing one constraint propagated through the rest. Parallel efforts ran the same year — CRAFT at MIT, DAC-1 at General Motors. Through the 1970s the field split into generic CAD and dedicated computer-aided architectural design (CAAD); in 1982 AutoCAD made vector drafting affordable on the personal computer and pushed CAD off the aerospace mainframe onto every drawing board on Earth. Architecture’s proof-of-concept came earlier and stranger: the Sydney Opera House shells were resolved on a Ferranti Pegasus in the mid-1960s, deriving a buildable surface rather than drawing one. Bilbao exists, as Gehry’s office has always said plainly, because CATIA — imported wholesale from Dassault aerospace — existed. The Heydar Aliyev Center has no canonical 2D drawing; only the NURBS model the drawings were sliced from.

In practice: In PAZ Atelier we treat the CAD file as the source code of the building, not a delivery format. A project begins with the parametric skeleton — the relationships, not the lines — and every drawing or BIM view is a downstream rendering of that one model. The discipline this enforces is small but absolute: if a number is not in the model, it is not in the project. For a Swiss studio this is also a survival move. As one recent industry piece on cloud CAD production argued, projects are getting more complex, deadlines shorter, and the skilled-worker shortage deeper — the studios that treat the model as a single authored source, rather than a pile of exported files, are the ones that absorb that pressure without drowning in coordination overhead.

Hack: This Hack teaches you to make the model a program, so changing one variable rebuilds the geometry — the lesson Sutherland was already demonstrating in 1963. The medium is OpenSCAD, where the .scad file is the model: no GUI to drag. The domain is Geometry. Open OpenSCAD, paste this, and move the slider on entasis:

// One program, one geometry. Change a number, rebuild the form.
h = 600; r = 60; entasis = 0.92; steps = 32; seg = 96;
module shaft() {
    for (i = [0 : steps - 1])
        translate([0, 0, i * h / steps])
            cylinder(h = h / steps,
                     r1 = r * (1 - (1 - entasis) *  i      / steps),
                     r2 = r * (1 - (1 - entasis) * (i + 1) / steps),
                     $fn = seg);
}
shaft();

The taper is no longer a line you redraw — it is a constraint the model honours. That is the whole of CAD in nine lines.

Here is the harder question, the one that does not show up in the feature list. CAD is moving in two directions at once: downward into the database, where BIM makes geometry a queryable, versioned artefact, and sideways into code, where OpenSCAD, CadQuery and Build123d return Sutherland’s original insight with proper version control and machine assistance on top. Both directions hand more of the authoring to the system. That is a real gain in speed and a real trade-off in agency: when the model finishes the relationship for you, you lose the productive pause where you would have decided whether the relationship was right. Save one move a day for the unaided version of yourself — model one joint by hand, reason through one tolerance without the optimiser — so the judgement stays yours when the stakes are high.