I am a building, and I have watched a lot of architects walk in carrying a laptop they do not fully trust. They open it, they show me a render some model spat out, and they ask the room — quietly — whether this is real work or a parlour trick. That question is the reason we are running the KI Sommer Camp 2026 again at PAZ Central, Niederdorfstrasse 77, in the heart of Zurich’s Niederdorf.

This is the third summer we have done it, and the shape that works is now clear: two days, on-site, hands on real architectural competition briefs — not slideware. We covered the 2025 edition on our own pages (pazacademy.ch/ki-sommercamp2025), and the spine carries forward: GenAI and LLMs, Python, and Grasshopper for computational geometry, taught the way you would actually use them on a Tuesday afternoon under deadline.

The signal worth naming is the shift in what the camp teaches. In 2024 the question was “can AI draw a façade.” By 2026 the question on every working desk is “can AI clean my project data, sort my competition references, and build me a small assistant I own — without leaking my client’s files into someone else’s cloud.” That last clause is not paranoia; it is the difference between a tool and a liability.

←TODAY: In June 2026 a Swiss practice can run a usable LLM workflow on a single laptop, on real Wettbewerb data, without a vendor subscription deciding what it sees. →3012: The buildings of the Zurich-3012 horizon will be designed by people who learned to instruct machines, not just prompt them. Fulcrum: Ownership of the workflow — local, legible, yours — is the one habit that survives every model generation that comes after it.

What two days actually buys you

The camp runs 09:00–16:30, on-site, in small groups — the same room PAZ uses for the Building System Specialist cohorts, three minutes from the Hauptbahnhof. Nora and Andrés lead it. You bring a laptop, or we provide one. The work moves through three honest stages: getting your project data into a shape an AI can read (cleaning, structuring, visualising); applying it — façade studies, option generation, reference triage; and building a small custom assistant you keep when you go home.

Where it touches our home turf, we say so plainly. Architects who work in Archicad will see where PAZGPT and the PAZ-BOX Archicad Add-Ons fit — AI chat and voice control of the model, GDL-aware automation — but the camp is platform-literate, not platform-locked. The point is the method, not the merch.

Atelier: The competition-brief format is the whole pedagogy. You are not learning “AI” in the abstract; you are sitting with a real Wettbewerb, a real site, and a real LOIN gap, and learning which task an AI genuinely shortens and which one it quietly ruins. That distinction is the skill.

Hack: This Hack teaches you to turn a messy folder of competition references into a structured table an AI can reason over — the AI/ML move the camp opens with. Drop your reference notes into a JSON list, then let a local script call a model to tag each one. The intention: stop pasting raw chaos into a chatbox and start handing it clean, owned data.

import anthropic
client = anthropic.Anthropic()  # key in your env, not the cloud UI
for ref in references:           # references = [{"note": "..."}, ...]
    msg = client.messages.create(
        model="claude-haiku-4-5-20251001",
        max_tokens=200,
        messages=[{"role": "user",
                   "content": f"Tag this AEC reference with 3 keywords, JSON only: {ref['note']}"}])
    ref["tags"] = msg.content[0].text

I have neighbours that went blind — control stacks locked in a cloud the vendor switched off in 2041, structures that can no longer read their own sensors. The same trap waits one floor down for your practice: a design workflow you rent but cannot open. The camp’s whole bias is toward the opposite — local, legible, fallback-able. Learn the move on terms you can keep.

Bring a current project and one stubborn data problem you have been avoiding. Sign up on our promo page — the KI Camp 2026 listing — and come build the assistant before the summer is over.