Generative AI in architecture is useful when it helps designers explore ideas faster, but it becomes weak when it replaces thinking with a pile of attractive images.
The SaSi Studio conversation with Mattia Santi is a good example of the better version of the topic. It connects AI to design experimentation, digital buildings, metaverse thinking and the practical question of how architects stay useful when tools move quickly.
Watch: SaSi Studio on generative AI in architecture
This SaSi Studio episode is useful because it looks at generative AI from inside practice: experimentation, digital work, design options and what architects still need to control.
Listen: the full SaSi Studio AI episode
The audio version gives you the longer conversation with Mattia Santi, including how SaSi Studio thinks about AI, architecture and digital design work.
What SaSi Studio adds to the AI debate
The original episode introduced SaSi Studio as a practice working across architecture, design and digital products. That matters because generative AI is not isolated from practice. It sits inside briefing, options, communication, visualisation and the way a studio competes for work.
Mattia’s point was not that tools such as Midjourney, ChatGPT or Stable Diffusion replace the designer. The useful point is that they can expand the design conversation when the designer knows what to ask and what to reject.
Where generative AI is genuinely useful
- Fast visual exploration before a direction is fixed.
- Testing mood, atmosphere, material direction and narrative.
- Helping clients understand options earlier in the process.
- Creating prompts, outlines and reference boards for internal discussion.
- Supporting smaller teams that need to communicate ideas quickly.
Where the hype breaks down
The danger is that generative AI can make weak thinking look finished. A beautiful image does not answer the brief, coordinate structure, solve access, meet planning constraints or explain procurement.
That is why architects still need to bring judgement: context, scale, regulation, material consequence, client priorities and the ability to turn a visual idea into something buildable.
How candidates can show AI properly
- Show the prompt or design question that started the exploration.
- Explain which outputs were rejected and why.
- Connect the image back to a real brief, user or site condition.
- Avoid presenting AI work as if it were fully resolved architecture.
- Make clear which parts are your judgement, drawing and design development.
A simple portfolio example
A useful AI portfolio page could show one brief, three generated directions, one selected route and the design development that followed. The important bit is not the tool. It is your decision-making.
- Brief: what did the project need to achieve?
- Exploration: what options did AI help you test?
- Selection: why did one route make more sense?
- Development: how did you turn the idea into architecture?
- Reflection: what did the tool miss that you had to solve yourself?
What practices should decide
Practices need simple internal rules. Who can use AI? What information can be uploaded? How are outputs checked? What belongs in a client presentation? What stays as internal exploration?
- For candidate context, read AI in Architecture Careers.
- For a related generative design episode, watch Algorithms and Aesthetics with Arka Works.
- For practical career next steps, browse live architecture jobs and resources.
Watch next: algorithms, aesthetics and AI tools
The Arka Works conversation takes the same AI theme into generative design, computational workflows and how candidates can make tool-led work legible in a portfolio.
Common mistakes
- Using generative AI only because it looks current.
- Showing images without process, constraint or judgement.
- Forgetting copyright, client confidentiality and attribution.
- Letting the tool define the architecture instead of the brief.
- Assuming every practice values AI in the same way.
Architecture Social view
Stephen’s recruiter view is that AI can help candidates stand out, but only if they can explain the work. A portfolio full of unexplained images may look busy, but it does not prove judgement.
Next step
Try one small AI experiment and write down the design judgement behind it. If you can explain what improved, what failed and what you checked manually, it becomes career evidence rather than a novelty. Keep learning through the Architecture Social podcast and related AI career resources.
Add a comment