AI in Practice: Four Perspectives on Where We’re Heading

RCA is responding to the growing use of AI in architecture through the establishment of a new AI Steering Group. The group currently comprises six members: myself, Mark Eden, Kirsty Pearson, Asu Amca, Bryony Preston, and Donald Kudangirana, each bringing a distinct perspective on AI and its practical implementation.

Our work is focused on four key areas: research, imaging and visualisation, project management, and operational workflows. As AI continues to evolve rapidly, members of the group regularly attend seminars and webinars led by architects and other professionals to understand how AI is being applied across the industry. We then share relevant insights and lessons with the wider RCA team.

I recently attended “AI in Architectural Practice: Real Projects and Real Impact” seminar, hosted by CMAP, that brought together four very different voices from across the built environment and technology sectors. What stood out wasn’t just the variety of tools being discussed, but the one message behind them. Across every presentation, the message was consistent: start with the problem, not the technology.

Here are the key takeaways.

1. AI Agents as “Virtual Clients” – BDG

Dan Stanton introduced the concept of AI Agent Preview, essentially a virtual client that allows teams to pitch, test and refine ideas before presenting to a real client.

The system uses Retrieval Augmented Generation (RAG) models, meaning it only draws from defined and verified source material. Every response is referenced. If the answer doesn’t exist in the dataset, the agent says so. This strict control of inputs is designed to minimise hallucination and increase reliability.

The key principle was simple: prioritise purpose over tools.

Before implementing AI, ask:

• What do you want AI to help with?
• What do you not want AI to help with?
• Where is the gap it should fill

The broader message was also important: look beyond architecture and learn from how other industries are using AI.

2. Physical AI and Microfactories – AUAR

Mollie Claypool shared AUAR’s work using “Physical AI” in containerised microfactories to support homebuilding.

Unlike conventional digital twins which often contain geometry only, AUAR’s system embeds real-world construction logic and building regulations into a “master builder model.” Parametric design is constrained by what can actually be fabricated in the microfactory to eliminate unbuildable outcomes.

Robotics respond to real-world variability, and the system creates a feedback loop between design and manufacturing. Builders can submit existing house types, which are then assessed for automation potential.

While perhaps less directly applicable to our day-to-day work, it represents a powerful example of AI operating beyond digital space, shaping how things are physically made.

3. Pattern Recognition at City Scale – RCKa

Russell Curtis presented an example of a Python-based, “vibe-coded” AI pattern recognition tool developed to identify small housing sites across London.

Using an InceptionV3 neural network and supported by a £1.2m Innovation Fund grant, the team analysed 2.1 million freeholds across all London boroughs. The AI identified small-site development opportunities with an estimated capacity of approximately 800,000 homes, around half of London’s housing target.

The system is built to handle extremely large Geographic Information System (GIS) datasets and was developed in collaboration with Blocktype and local councils.

An interesting discussion emerged around the “black box” nature of neural networks. These systems don’t understand what they are seeing in human terms. The question becomes: does understanding the internal logic matter, if the outcome is reliable and useful?

The key takeaway here: find a clearly defined problem, then work backwards to build a tailored solution.

4. AI Embedded in Practice Management – CMAP

Ben Jervis focused on AI integrated directly into CMAP Intelligence, targeting everyday operational ‘pain’ points:

• AI-powered fee estimation
• Automated resourcing
• Predictive email filing
• Project insights for improved decision-making

He framed AI adoption around a deceptively simple question:
What is the one task you never want to do again?

Rather than pursuing AI for its own sake, the focus was on removing friction from workflows and freeing up time for higher-value work.

Reflections

Across four very different applications – virtual clients, robotic microfactories, urban-scale pattern recognition, and automated practice management, a consistent theme emerged:

• Control the data.
• Define the purpose.
• Solve a real problem.
• Then build the tool.

AI is not a singular technology but a spectrum of approaches. Its value lies not in novelty, but in precision: identifying a gap and designing a system specifically to fill it. Moving forward, it will be important to share this knowledge across the practice and consider how RCA can adopt AI in a focused, strategic, and purposeful way.