The future of AI in design will be decided less by the software than by the people who sign the drawings. For construction managers, this is both a warning and an opportunity, says Eleni Stefania Kalapoda
Most conversations about AI in construction begin with what the tools can do: generate layouts, test compliance, draft specifications, coordinate models in minutes rather than weeks. In England, a more important question comes first: when something goes wrong, who answers for it?
The Building Safety Act has already answered this. Since October 2023, the Building Regulations have named five dutyholders: the client, the principal designer, the principal contractor, designers and contractors, each of whom must be competent for their role.
For higher-risk buildings, the Building Safety Regulator must be satisfied at Gateway 2 that the design complies before a spade enters the ground; at Gateway 3, the principal designer and principal contractor declare that they have met their duties before a single resident moves in.
Running through it all is the golden thread: an accurate, living record of how the building was designed, built and changed. Nowhere does that framework mention algorithms, and it does not need to. AI changes how design information is made; it does not change who is responsible for it. A facade option generated overnight, a specification clause drafted by a language model, an automated clash report: each carries the same weight as work produced by hand, because a named person will rely on it, sign for it, and answer for it if it fails.
The future of AI in design will be decided less by the software than by the people who sign the drawings. For construction managers, this is both a warning and an opportunity. The warning is that machine-generated content can be fluent, persuasive and wrong, and can flow into a project’s information faster than anyone can check it.
Five practical steps
The opportunity is that the Act has already built the scaffolding to govern it. Five practical steps would put that scaffolding to work:
- Record provenance at the source. Every drawing, model element, calculation, or specification produced or substantially shaped by AI should declare it in its metadata: which tool, which version, which inputs, and who prompted it. Most common data environments can already carry this information. A golden thread worth trusting must show not only what was decided, but also how the information behind each decision came to be.
- Give every output a name. AI cannot be a dutyholder. Before sharing any AI-assisted content, a competent person should review it, accept it, and put their name to it, exactly as they would for a junior colleague’s work. Responsibility matrices and information delivery plans should make that ownership explicit, rather than leaving it buried in a software licence.
- Anchor review to the gateways. AI is used most intensively at the start of a project, in option studies, massing and early compliance checks, which is precisely the information that feeds a Gateway 2 application. Teams should set formal review points before AI-assisted work is fixed into the submission, with the principal designer’s sign-off recorded alongside it.
- Treat every regeneration as a change. Higher-risk buildings already operate under rigorous change control: controlled changes recorded, notifiable changes reported, and major changes approved before work proceeds. AI can redraw a design in seconds, which makes it tempting to treat revisions as trivial. Change control plans should state plainly that a regenerated output is a change to be assessed, recorded and, where required, notified, however minor it appears.
- Build it into appointments and competence. Clients should ask how their designers and contractors use AI, and appointments should set clear expectations on disclosure, verification and data handling. Competence frameworks must evolve as well. Knowing when to question a machine’s answer is now part of doing the job.
Don’t slow down
None of this is a reason to slow down. Used well, AI can make buildings safer: catching coordination errors earlier, testing fire and environmental scenarios more thoroughly, and keeping the golden thread more complete than any manual process could. But that promise rests on trust, and trust rests on accountability.
If the past decade has taught our industry anything, it is that information failures become safety failures. We have spent years rebuilding a culture in which someone is always responsible. As AI reaches into every stage of design, our task is to carry that culture with it, so that no matter how a drawing is made, a competent person always stands behind it.
Eleni Stefania Kalapoda is an ARB-registered architect and urban designer. She has worked with the likes of the Renzo Piano Building Workshop and Foster + Partners. She is a 2026 Women In Construction & Engineering Awards finalist.
