What (not) to do with GenAI?

A Revit model with drawings in under 30 minutes. A visualisation in under a minute. Feedback within seconds. Artificial intelligence lets us produce and test designs at a scale never seen before, and its potential can no longer be ignored.

GenAI is not a fringe tool for tech enthusiasts. It is gradually changing the way we design: from the architect of individual solutions to the architect of systems that generate, compare and evaluate countless alternatives.

The series aims to open a wider discussion of this shift at the Faculty of Architecture of the Czech Technical University (FA ČVUT). SPECT Community lecturers and their guests draw on their experience of teaching at FA ČVUT, UMPRUM and ARCHIP and of practice at MVRDV and CodingARCH. They will show how GenAI is changing the design process, where its risks and limits lie, how to work with it so that people stay in control, and they will ask what, and how, we should teach future architects.

The series has two parts. The first, a set of short talks closing with an open discussion with the audience, raises the wider questions of GenAI’s impact on education and the pitfalls that students and teachers face today. Saman Saffarian, AI ambassador at FA ČVUT, will add his own observations.

The second part, led by Fredy Fortich, an AI researcher at MVRDV, shows how the technology is integrated into each phase of the design process within the office. In two workshops, participants then try current tools and explore, on concrete examples, the potential presented in the talks.

The series does not aim to give one definitive answer. It is meant to create room for shared learning and common conclusions that can shape how architecture is taught next.

Part one: GenAI in teaching

11 November 2026 from 15:00, 3 hours in total. Experience from practice and from teaching at FA ČVUT, UMPRUM and ARCHIP. The talks are in Czech.

• 15:00 Šimon Knettig: From designer to critic? Where GenAI is taking the profession
• 15:40 Lukáš Kurilla: Speed is only one of the variables; quality comes from the system
• 16:20 Šimon Prokop: From wires to dialogue: procedural modelling in the age of GenAI
• 17:00 Guest: Saman Saffarian, AI ambassador at FA ČVUT (title to be announced)
• 17:30 Moderated discussion with the audience

Each talk lasts 30 minutes, with a 10-minute break between talks.

Šimon Knettig: From designer to critic? Where GenAI is taking the profession

When Christopher Frayling formulated the concept of research through design in 1993, he defended the design process itself as a way of gaining knowledge, and with it the academic relevance of design work. But what happens when creating one design no longer takes weeks or months, but hours or minutes? When we can generate dozens or hundreds of alternatives in a moment?

Where does the architect’s work go then? Will we spend more time comparing alternatives than creating them? How will we know which one is better? By which metrics will we evaluate them, and by which values will we generate them in the first place?

GenAI changes not only the speed of design, but the very centre of the architect’s work: from designing individual solutions to designing the system that generates and evaluates them. From the architect of solutions to the architect and critic of the system.

Using an experimental studio at ARCHIP as an example, we will show how students coped with this shift, what the new approach made possible for them, where they hit its limits and what questions it raises for teaching architecture. The talk sets out the theoretical and practical frame for the seminar that follows.

Lukáš Kurilla: Speed is only one of the variables; quality comes from the system

Most debates about GenAI in the design process revolve around automation and speeding up work. But speed is only one of the variables in the equation of success. Just as important is quality that we can demonstrate. This is where the double diamond comes in, a method framework created by the British Design Council in the early 2000s. In the spirit of design thinking, it was meant to help find the right solution for a well-defined problem, that is, to deliver a design that meets users’ real needs. But what is a good solution, and how do we recognise it among generated alternatives? By which criteria do we choose, and who sets them?

The extended double diamond we present connects design thinking with computational thinking. To each qualitative step (ambitions, concept, design) it adds a measurable counterpart: requirements, KPIs, performance. We then use generative AI and algorithmic modelling not to browse references, but to understand the problem more deeply and to test an incomparably wider field of solutions. The result is not just a faster design, but a demonstrably better one. The way of thinking changes too: instead of a single solution, we design the rules, criteria and values from which solutions emerge.

On projects from teaching and practice, we will show how the framework works in the concept phase as well as in rapid prototyping and testing, why computational thinking is key right now, and what skills a regular user, an advanced user and someone who builds their own tools will need. The next talk builds on this method framework with a look at specific tools.

Šimon Prokop: From wires to dialogue: procedural modelling in the age of GenAI

Almost twenty years ago, Grasshopper opened computational design to people who do not want to program. Instead of writing code, you assemble components and connect them with wires. Language models now push this boundary further. They write a script component in Python or C#, explain someone else’s definition, help find an error in a data tree and, thanks to protocols such as MCP, are starting to work with Rhino and Grasshopper directly.

With live demos, the talk follows what this changes. Manual wiring fades into the background and the work shifts to breaking a problem into steps, describing the logic and checking the result. Unlike image generation, the output stays a readable and editable definition: rules you can keep working with. We will also name the risks: definitions their author does not understand, code that breaks at the first change of the brief, and the illusion that parametric thinking is no longer needed.

With examples from teaching and practice, we will show when it pays to let GenAI write, when to use it only as an adviser and when it is faster to reach for components. The talk adds a view of everyday work with tools to the previous talks.

Guest: Saman Saffarian

AI ambassador at FA ČVUT.

Part two: GenAI in practice

3–4 December 2026. Experience from MVRDV. Lecturer: Fredy Fortich, architect, engineer and GenAI researcher at MVRDV. The talk and workshops are in English; the detailed content will be announced.

• Talk AI across the office design workflows (1 hour)
• Workshop 1 Design iteration through images and diffusion models (4 hours)
• Workshop 2 3D modelling using LLMs and MCP (4 hours)