by Gary Barley, Technical Director – Digital, Pascall + Watson
Caution should not be confused with resistance
The proverb “time and tide wait for no man” feels relevant to the pace of AI development. It is hard to look at the news or attend an industry event without encountering a claim about what AI may change, improve, or disrupt.
Public debate rightly concerns regulation, development pace, guardrails and oversight. But AI is already part of everyday working life, it’s embedded in the software we use and increasingly expected to be used by our clients and colleagues.
We should not resist AI but neither adopt it foolishly. AI can reduce repetitive work, support research and create more time for design thinking and problem-solving.
Architects must remain cautious as we are accountable for advice, information and decisions, regardless of tools that assist us. AI may support the process but cannot assume the professional responsibility.
Enabling safe and managed exploration of AI
Within our practice, there is healthy curiosity about AI. Colleagues are already personally exploring tools to generate early visual studies, summarise information, support research and reduce repetitive tasks. A blanket ban may appear safest but would drive experimentation into personal accounts and public platforms reducing control over data, security, and output quality. Unrestricted access was not appropriate either, where privacy, confidential project information and intellectual property are concerned.
Our response was to provide secure, managed access to language and image generation tools, alongside a controlled sandbox for testing new technologies. This gives colleagues a risk-free route to explore value, share learning, and understand AI’s limits.
Start with the low-risk applications
Responsible experimentation does not need to begin with the most complex or high-profile application of AI. It is often better to start with practical tasks where the consequence of an error is lower, the benefit is clear and a manual review is straightforward.
For architects and designers, suitable early uses might include producing visual studies from sketches, model views, or reference material, to summarising meeting transcripts, searching approved internal policy or project documentation and assisting with coding, scripting or data cleaning. These uses require appropriate access controls and a review by someone with relevant knowledge.
AI can still be used on higher-risk work, but only where a suitably qualified person can check the result and take responsibility for it. Responses relating to the building regulations, technical specifications and contractual communications, if inaccurate or incomplete, could have consequences.
Accountability remains with the Architect
AI can draft, summarise, translate, generate and analyse rapidly. It can also produce convincing but inaccurate, incomplete, or context inappropriate information. It may omit qualifications, reflect bias, limitations, or lack a reliable source trail.
For this reason, oversight by a real person, cannot be an afterthought, it must be built into the workflow from the outset.
Architectural practices are not strangers to this. Work by a recently qualified architect under mentorship is not issued without appropriate review. An experienced colleague checks it is accurate, coordinated, and fit for purpose. This is normal professional responsibility and quality assurance.
The same principle applies to AI assisted work, it requires authorship, review and sign-off. Technology may assist the process, but the responsibility remains with the architect.
Build confidence through learning and freedom to ask questions
Confidence in AI varies within an architectural practice. Some colleagues may be enthusiastic early adopters, eager to test new capabilities without fully understanding limitations. Others may feel uncertain about their knowledge, question AI’s relevance or worry about its implications for their future in the profession.
There is no single route to confidence. Some people learn through hands-on experimentation, others benefit from guidance, demonstrations, structured training or examples they can revisit. Practices should offer a range of learning opportunities rather than expecting one format to suit everyone.
Informal “lunch and learn” sessions can create space to share workflows, discuss failed experiments and identify when a tool was inaccurate or unsuitable. These less polished examples often offer the most valuable learning.
Ultimately, responsible experimentation depends on openness. People should feel able to ask questions, disclose AI use and seek advice before relying on an output. A culture where people can say “I am not sure” is essential.
A responsible approach will provide better outcomes
Responsible experimentation gives practices of all sizes a practical route to develop their AI capability without losing sight of the principles that define who they are. It allows people to explore useful tools, test new workflows and learn from experience, while keeping design decisions and responsibility with the people doing the work
This approach is not only about technology, but also about bringing people with you. It means recognising different levels of confidence, giving people time to try the tools, raise concerns and share their knowledge.
The practices that will benefit most from AI aren’t necessarily those that move fastest or adopt and test every new tool first. It is more likely the ones to think strategically, engage their staff and retain control over how technology is used.
