Artificial intelligence in architecture is entering a new phase. What began only a few years ago as an explosion of AI-generated imagery is rapidly evolving into something far more consequential: intelligent systems capable of operating design software, generating three-dimensional models, evaluating technical requirements, and participating directly in architectural workflows. For architect and designer Tim Fu, this transition signals not simply another technological upgrade, but a fundamental restructuring of how architecture may soon be practiced.
Fu joined Hamid Hasanzade, founder and CEO of Parametric Architecture and PAACADEMY, for a new conversation on Designing Forward, the continuation and evolution of the former PA Talks podcast. Under its new identity, Designing Forward expands the conversation around architecture, design, technology, and culture, focusing on the people and ideas shaping what comes next. In this episode, Hasanzade and Fu discussed the arrival of agentic AI, Astra, Studio Tim Fu’s growing portfolio, the launch of STF Intelligence, and an increasingly urgent question: what remains uniquely human when AI can participate in almost every stage of architectural production?
From AI Imagery to AI-Native Architecture
Tim Fu emerged as one of the recognizable figures of architecture’s early generative AI movement after leaving Zaha Hadid Architects and establishing his own London-based practice, Studio Tim Fu. What initially gained attention through experimental AI imagery has since developed into a practice of around ten people working on increasingly ambitious real-world commissions.
Among the studio’s current projects is a substantial masterplan in the Dominican Republic encompassing approximately 70 villas and multiple architectural components, alongside resort development work in Thailand, an ongoing project in Switzerland, and a high-end wellness project in London. For Fu, however, the evolution of the studio is happening simultaneously on two fronts: architecture and technology.
The distinction is important because Studio Tim Fu is no longer approaching AI merely as software added to a conventional architectural workflow. Instead, the studio is restructuring its processes around the capabilities of increasingly autonomous AI systems. Fu describes this transition as the arrival of an “agentic era,” in which AI moves beyond producing content and begins taking actions within the tools architects already use.
Astra and the Arrival of Agentic AI in Architecture
A significant portion of the conversation focused on Astra, which Fu considers one of the most disruptive developments yet for AI in the AEC industry. His research team began rapidly experimenting with the model following its release, investigating how its capabilities could be incorporated into the studio’s existing AI infrastructure.
Rather than attempting to build foundational AI models independently, Fu explained that his team works on top of frontier models, creating what he describes as scaffolded agents specialized for architectural decision-making. These systems incorporate knowledge about architectural processes, design logic, industry pipelines, and software, while benefiting from improvements in the underlying general-purpose models.
The implications become particularly significant when AI is connected directly to professional design environments. Fu says his team is developing internal plugins that enable agents to interact with Rhino, AutoCAD, and Revit, allowing architectural models, drawings, renders, and design decisions to be produced through natural-language instructions rather than exclusively through manual software operation.
For Fu, this represents a major shift because AI is beginning to enter precisely the areas traditionally occupied by junior architects and designers: building models, resolving geometry in three-dimensional space, testing alternatives on a site, applying materials, producing iterations, and comparing design options.
“The fear is real,” Fu acknowledged during the conversation, while emphasizing that his response has been to ensure everyone within the studio actively uses AI rather than treating it as an external threat. The more consequential challenge, he argued, is that architecture practices now need to reconsider what their design process should look like when many previously manual tasks can increasingly be delegated to intelligent systems.
Could AI Take Over Architectural Documentation?
The transformation does not stop with conceptual design. Hasanzade pushed the discussion toward one of architecture’s most labor-intensive territories: documentation, compliance, and technical delivery.
Fu imagines a future involving what he calls a “council of agents,” in which specialized AI systems could evaluate different aspects of a project. One agent might understand local building codes and examine drawings for compliance, while another evaluates sustainability, and additional agents address other specialized technical requirements. Their assessments could ultimately feed into a coordinating system capable of balancing competing criteria.
This possibility shifts the AI conversation away from spectacular visualization toward the less glamorous but economically significant work underlying architectural practice. Fu believes many later-stage activities could eventually become highly automated because they consist of requirements that can be systematically analyzed and resolved.
What remains more difficult is creative judgment. Yet even here, Fu is reluctant to draw permanent boundaries around what machines will never accomplish. He described creative decision-making as perhaps the “last bastion” of human involvement, before immediately questioning how long that distinction can survive.
“The only thing I’ve learned from this entire journey is that you shouldn’t underestimate what AI can’t do,” Fu said.
Architecture Beyond the “Instagrammification” of AI
The discussion also confronted another consequence of generative design: an enormous volume of seductive architectural imagery created primarily for digital consumption. Hasanzade questioned whether AI’s ability to produce increasingly spectacular projects has necessarily resulted in better spaces.
Fu sees greater historical continuity than rupture. Architectural magazines, experimental movements, and groups such as Archigram previously offered platforms for provocative and sometimes deliberately unrealistic architectural propositions. Social media accelerates that tradition dramatically, allowing speculative ideas to circulate at unprecedented speed.
The problem emerges when production becomes detached from intention. Fu acknowledged the prevalence of “AI slop” and described a broader “Instagrammification of architecture,” in which projects can become optimized for immediate attention and virality rather than enduring architectural value.
For Fu, the real test comes afterward: whether attention can be translated into meaningful architecture. A compelling image may initiate a conversation, but architecture ultimately has to create a sense of place capable of surviving beyond the screen.
Why Parametric Thinking May Become More Important, Not Less
Perhaps counterintuitively, Fu does not believe the rise of natural-language interfaces makes computational knowledge irrelevant. His years working with parametric design have instead become increasingly valuable because communicating effectively with AI requires understanding how a problem should be structured.
Rather than simply instructing an agent to “design a bridge,” an architect who understands procedural and parametric logic can describe relationships between structural loads, beam distribution, tension and compression members, geometrical constraints, and performance requirements. In Fu’s view, understanding parametric, procedural, and algorithmic thinking allows designers to orchestrate AI with substantially greater precision. tim-hamid.txt
That shift also informs how Studio Tim Fu evaluates talent. As repetitive production becomes increasingly automated, Fu believes critical thinking, experimentation, orchestration, and creative problem-solving become more valuable. His advice to emerging architects is consequently straightforward: experiment with the technology, develop original workflows, and demonstrate what can be achieved rather than merely listing software proficiency.
From Studio Tim Fu to STF Intelligence
The clearest indication of where Fu believes architecture is heading may be his decision to move beyond architectural practice itself. During Designing Forward, he revealed that a separate technology company, STF Intelligence, had recently been incorporated as a spin-off from Studio Tim Fu.
The new venture is intended to focus specifically on developing AI tools for architects, eventually bringing some of the studio’s internally developed technology to a wider audience. Fu sees an opportunity between two worlds: technology companies often possess extraordinary engineering capabilities without the depth of knowledge embedded in architectural practice, while established architecture firms possess that expertise but may struggle to adapt their organizations rapidly enough to technological change.
Studio Tim Fu has the unusual advantage of having grown during the AI revolution itself. Its structure can therefore change alongside the technology rather than requiring decades-old processes to be dismantled first.
That adaptability may ultimately be the defining theme of this Designing Forward conversation. AI’s architectural impact is no longer confined to generating provocative images of buildings that do not exist. It is beginning to enter modeling, documentation, compliance, optimization, education, and eventually construction workflows. The question facing architecture is therefore shifting from whether AI belongs within the profession to something much more difficult: how should architects redefine their value when intelligent systems can participate in almost everything they once considered architectural work?
Prefer to listen? Tune in to Episode 73 of Designing Forward on Spotify and hear the full conversation with Tim Fu on agentic AI, Astra, and the future of architecture.
