Is there a ChatGPT for architecture?

AI now writes emails and code, so people reasonably ask whether it can design buildings. Not yet, and not without architects.

What people picture

Mention AI for architecture and many people imagine a digital super-architect that can:

• Produce fully compliant floor plans instantly
• Generate planning applications automatically
• Create photorealistic renders with no input
• Solve any design problem at the click of a button

It is an appealing idea. For now it is mostly fantasy.

What AI can do today

We are not handing design over to machines, but AI already helps us work through design problems. We use AI tools to:

• Summarise complex briefs
• Explore early feasibility options
• Test spatial layouts quickly
• Pre-screen compliance data
• Help clients see their choices more clearly

We treat these tools as an aid to design thinking. The thinking stays ours.

The tools available now

The range of AI platforms for architects is growing. It includes:

• Spacemaker AI for site optimisation
• Hypar and TestFit for generative layouts
• Veras, Ark AI and LookX for visual ideas and image generation
• Custom GPTs for planning policy lookups, code checks and BIM workflows

These tools are fast. They do not replace an architect’s judgement, which matters most on complex projects where context, character and people are at stake. We set out why in a better way to design.

Where AI stops

AI can automate some outputs. It cannot:

• Read the finer points of what different stakeholders need
• Balance beauty and function
• Weigh competing constraints against each other
• Interpret the emotional and social context of a space

Those are the things that make architecture meaningful, and they do not automate easily.

How we use it

We use the best tools available, with a clear purpose and people directing them. We look for places where AI can speed up the process without weakening the result. The aim is clearer information for clients, quicker insight and better design decisions. The same thinking runs through our work on MMC and DfMA.