Fifteen Years of KI in the Office: PAZ Back-to-Work Camp, 23 September
A one-day PAZ Academy session on 23 September 2026: fifteen years of AI and computation habits for architecture offices returning to work. Bring one real task.
When Dr. Mikhael Johanes, a Design++ postdoctoral fellow at ETH Zurich, closed our KI Summer Camp — the 2–4 July 2025 edition — with a keynote on machine learning in architectural design, the point that landed hardest was a timeline. Most of the methods now reshaping the drawing board are old research — decades of it — that has only recently become cheap enough to run inside a working Büro. That gap between when a method is proven and when it is affordable is the whole story of computation in practice, and it is why we keep a camp on the calendar.
So, the signal: on Wednesday, 23 September 2026, we run a one-day session at PAZ Central — the camp we have named PAZ-Working-with-KI-since-15-Years. It is a back-to-work day for offices coming off the summer, built around a claim we can actually back: we have been folding AI and computation into architectural work for fifteen years, and the useful part is the habits that survive the hype cycle rather than the hype cycle itself.
←TODAY: In 2026, an office can run a retrieval model over its own past briefs on a laptop before lunch. →3012: In the Zurich-3012 we write toward, the schools that kept teaching fundamentals still have students; the ones that sold only tools were forgotten. Fulcrum: The tool changes every year; the practice of asking a clean question of your own data holds.
Why fifteen years matters
The number is not decoration. The PAZ library that seeds our Building System Specialist title — the five-month program, six hours a week, that has certified 39 specialists — draws on more than fifteen years of accumulated scripts, templates and tools. That is what a camp like this trades in: a spine of reusable moves rather than a new model to be dazzled by. When we show an office how our own PAZ-BOX brings Archicad automation within reach of an AI chat prompt, the demo only means something because there is a decade and a half of workflow logic underneath it.
The camp covers the ground our summer editions did — GenAI and LLMs, Python and Grasshopper for computational geometry, cleaning and visualising project data, building a small custom AI assistant — compressed into a single day and pointed at the September return-to-work reality: a full inbox, a stalled competition entry, and no time to learn a tool from scratch. As our KI Summer Camp 2025 write-up records, that curriculum runs on real architectural competition briefs, not toy examples.
What it looks like at the desk
Picture the first Monday back. Three competitions half-read, a model that needs cleaning, and the quiet suspicion that half the week will vanish into re-typing things a script could hold. Our summer camps have run CHF 1,790–2,190 and filled PAZ Central at Niederdorfstrasse 77 from 09:00 to 16:30 precisely because that suspicion is correct — the returns compound once an office stops treating AI as a novelty and starts treating it as plumbing.
Atelier: The shift a returning office feels this autumn is that AI has moved from the demo table into the daily file. The one move to make Monday: pick a single repetitive task from last week — renaming layers, extracting quantities, drafting a competition summary — and write down the exact prompt or short script that would remove it, then trial it once on real project data before you scale it anywhere.
Hack: Turn your last ten competition briefs into a searchable memory before you write the next one. The move is to embed them once and query them cheaply, so the office’s own history answers before a generic model does. A few lines of Python get you a working retrieval store you can grow all year.
from openai import OpenAI
client = OpenAI()
briefs = open("competition_briefs.txt").read().split("\n\n")
vecs = [client.embeddings.create(model="text-embedding-3-small", input=b).data[0].embedding for b in briefs]
print(len(vecs), "brief vectors ready — now cosine-match a new brief against them")
Swap in a local model if your data stays in-house; the pattern is identical. That is the whole camp in miniature — one clean question, asked of your own data, answered fast.
If you were at our July editions, you already know the rhythm: the room goes quiet, someone’s script finally runs, and the sound a room of architects makes when it does is the reason we keep hosting. This edition is the same craft, tuned for the back-to-work week. The full session outline and the sign-up path live on our camp page — bring a laptop and one real task you want gone by October.
PAZ Kaffi · multidisciplinary editorial, led by PAZ Academy