AI in the Architecture Office: Where It Helps and Where It Gets Dangerous

An architecture office lives on design. But it earns a surprising share of its money through text: project descriptions, explanatory reports, emails to clients, tender documents, portfolio pieces. None of it counts as creative work under Germany's fee schedule for architects (HOAI), all of it costs evenings.

That share of writing is exactly where AI changes something in an office. Not in the design. Not in the structural engineering. In the language surrounding it.

What an office actually writes

Try counting for a week. A client's inquiry that needs a detailed reply. The project description for the website. The explanatory report for the building permit application. The email to the specialist engineer that needs to stay friendly even though the deadline is blown. The tender texts for forty line items.

In an office of five people, that adds up to ten to fifteen hours a week. That's a third of a full-time position that designs nothing at all.

Where AI reliably helps

Project descriptions for the website and portfolio

You designed the building, you know every angle of it. What's missing is the text that conveys to a layperson why the arrangement of rooms is clever. Give the AI the key facts and your thoughts in keywords, and it builds a flowing text out of them.

Client communication

Clients aren't experts. The amount of explaining required is enormous, and it repeats itself. Why the building permit is delayed. What a change order means. Why the change costs more than it appears to save.

AI translates technical language into everyday language, and it does it patiently. You write out the situation in your own words, ask for a version for someone with no construction knowledge, and get back a text you just need to review.

Tender and specification texts

Caution is warranted here, but the benefit is significant. For the wording of a line item, AI can supply a framework. For technical accuracy, standards references, and completeness, you're responsible. Always.

Internal organization

Meeting minutes from keywords, checklists for site supervision, onboarding materials, job postings. All texts with no project data attached, all harmless, all faster.

Where AI fails in the architecture office

And now for the more important part. There are three areas where AI isn't just useless, it's dangerous.

First: technical planning. Structural engineering, building physics, fire protection, energy calculations. An AI can compute, but it can't be held liable. And it doesn't notice when its assumption was wrong.

Second: legally binding texts. Contracts, change orders, defect notices. The exact wording decides claims. A change order that the AI phrased more elegantly can cost you money, because one word shifted its meaning.

Third: anything containing personal data. The client's name, the property address, contract sums. None of that belongs in a general-purpose AI tool.

A workflow that works

• Gather your thoughts in keywords, with no names or addresses.
• Tell the AI who the text is for: client, authority, website, colleague.
• State the tone and length you want. Without that, it guesses.
• Check every technical detail in the result against your documentation.
• Cut any phrasing that promises something you don't want to guarantee.
• Insert names, addresses, and figures only in the finished document.

What remains for you

AI isn't going to produce a design in an architecture office, check a building permit application, or carry any responsibility. It's not going to replace technical planning either, and anyone who claims otherwise has never actually seen an office from the inside.

What it can do is thin out the layer of writing that sits on top of your work. Four to six hours a week where you're planning instead of wording things. That's a modest promise. It's also one that actually delivers.