Somewhere between the last out-of-office reply and the first Monday of autumn, a quiet thing happens in every architecture office: the AI tools you meant to learn over the break are still exactly where you left them, unlearned. That gap — between the headlines you scrolled on the train and the workflow on your screen — is the whole reason we built the KI Back 2 Work Camp, running 23–24 September 2026.

We named it plainly on purpose. This is not a conference about the future of AI; it is two days about the first working week after summer, when a Büro has to decide what it will actually do differently. Künstliche Intelligenz only earns its place at a desk when it survives contact with a real deadline.

←TODAY: On 23–24 September 2026 we host the KI Back 2 Work Camp in Zurich — two days to convert a summer of AI headlines into repeatable desk habits. →3012: The offices that last teach their tools to the next hire instead of hoarding them in one person’s head. Fulcrum: An AI skill only compounds when the whole team can repeat it Monday morning, not just the one who sat the workshop.

The lineage behind a two-day camp

This camp sits in a line PAZ readers already know. We ran the KI Summer Camp from 2–4 July 2025, three days at 09:00–16:30 out of PAZ Central, Niederdorfstrasse 77 in Zurich, working real competition briefs with a closing keynote from Dr. Mikhael Johanes — a Design++ postdoctoral fellow at ETH Zurich — on machine learning in architectural design. What we learned there shaped this one: architects do not need another tour of generative tools, they need a rehearsal of the moment a script finally runs and the room goes quiet. The Back 2 Work format compresses that into a focused two-day return-to-desk rhythm — the European autumn upskilling window, used honestly.

The through-line is the same as our Building System Specialist track — the five-month program we run at six hours a week, drawing on 15+ years of PAZ scripts and templates: staged, not scattered. As we’ve written before on the KI Back 2 Work Camp page, the point is to leave with a system you can run, not twenty fragments that never connect.

What lands on your desk

For a practising architect, BIM specialist, or engineer, the practical value is narrow and specific: less time on the repetitive parts of the model, more on the parts that need your judgement. That means AI that reads a competition brief and surfaces the clause you forgot, tools that talk to your BIM data instead of around it, and the beginnings of Archicad automation you can maintain yourself — including where a companion like PAZ-BOX lets you reach Archicad through chat and voice rather than clicking through every dialog.

Atelier: The offices that come through an AI adoption well share one trait — they pick a single workflow to change, not ten. This week’s Monday move for your Büro: take one recurring task (renaming exports, checking property sets, drafting the Wettbewerb cover text) and write down, in one sentence, the outcome you want an assistant to produce. Bring that sentence to the camp; it is worth more than a wishlist of software.

Hack: Turn your last competition brief into a memory your desk can actually query. The move: embed the text once, then match questions against it instead of re-reading twenty pages. Three lines of Python get you there:

from sentence_transformers import SentenceTransformer
model = SentenceTransformer("all-MiniLM-L6-v2")
briefs = open("wettbewerb_2026.txt").read().split("\n\n")
vecs = model.encode(briefs)   # one vector per paragraph
# cosine-match a new question against vecs to pull the exact clause

That is the whole idea in miniature: your office’s own documents become searchable by meaning, not by Ctrl-F. On day two we build this out against real BIM and IFC data, so the assistant answers from your project, not the open web.

A fair caution, because we won’t oversell it: an AI that reads your briefs can also quietly encode a bad assumption and repeat it across every future model. The discipline the camp insists on is the boring one — you check what it produces before it ships, every time. A school that only sells its tools dies; a school that teaches you to audit them is doing its job.

If your September plan is to finally close the gap between what AI can do and what your desk does, block the 23rd and 24th, read the agenda on our camp page, and come with that one sentence written down.