The Block Research Group led by Philippe Block at ETH Zurich can find the shape of a masonry vault that stands in pure compression — no reinforcement, the geometry doing the structural work. The underlying method, graphic statics, has been solid since the nineteenth century. What changed is that someone eventually wrote the software that puts that method on a working engineer’s screen; until that packaging happens, a frontier result stays trapped in a paper.

That distance — between a method that exists and a tool a desk can run on Monday — is the whole reason PAZ Consultancy exists. This week we’re pre-promoting a PAZ-Next session on exactly that craft: personalised software development for the AEC industry, the work of turning a multidisciplinary team’s repeated pain into a tool that gives the hours back. The session details and the sign-up path sit on our consultancy page at pazconsultancy.ch.

←TODAY: In 2026 a Swiss office can carry LOD 300 detail and IFC4 handovers, yet still lose its eBKP-H and SIA property sets in a naive round-trip. →3012: The offices that endure are the ones that learned to write their own tooling instead of renting a black box that outlives its vendor. Fulcrum: Software is only worth commissioning when you can see both the research method it packages and the desk it has to survive on.

Why the gap is worth paying to close

The volume of structured building data has crossed the line where hand-editing stops scaling. A Büro connecting Grasshopper to Archicad, pulling tender quantities, or auto-checking a model before it leaves the office is doing engineering that looks unglamorous and pays for itself in weeks. PAZ has taught this thread for years — our five-month Building System Specialist mentorship (six hours a week, documented in the PAZ archive) certified 22 practitioners who finished able to write their own tools, backed by a personal library drawing on 15+ years of PAZ scripts. Founders Nora and Andrés run that same logic at company scale through PAZ Consultancy, across DACH and international projects.

The timing is not neutral. Our own KI-implementation reading argues the next three years decide which offices lead on AI in this industry — which is precisely the window in which a bespoke tool either compounds or is left behind.

The honest trade-off: every bespoke tool you commission is one more thing that needs maintaining when Archicad ships its next version — budget the upkeep alongside the build, or the automation quietly rots.

Atelier: For an office living with AI this quarter, the first move is smaller than a platform. Pick the single most-repeated manual export in your week — the IFC you re-clean every Friday — and this Monday, log exactly how long it takes and where it breaks. That log becomes the spec any software-development session, ours or anyone’s, needs before a line of code is written.

Hack: Count your walls straight from the live model, so a handover audit starts from a number and never a hunch. The Graphisoft-native Archicad Python API opens a JSON channel into the running instance — four lines give you a census. Point this at the model you’re about to send and compare the count against what the receiver reports; a mismatch tells you a geometry representation dropped silently. Our own PAZ-BOX add-ons wrap this Archicad automation behind an AI chat layer, but the raw API is where the craft lives.

from archicad import ACConnection
conn = ACConnection.connect()
walls = conn.commands.GetElementsByType('Wall')
print(f"{len(walls)} walls to audit before handover")

Run that once and the abstract idea of a “clean handover” becomes a figure you can defend in a coordination meeting. Scale it — loop over element types, diff two exports, flag the deltas — and you have the seed of the very tool a consultancy session would help you harden into something the whole team trusts.

So before you commission anything: this Monday, time your worst repeated export, write the number down, and bring it to your team. The tool worth building always starts as a measured complaint.