The Duty to Observe, Revisited

Photo by Olek Buzunov on Unsplash

One aspect of professional practice I miss in retirement is the architect’s obligation to conduct periodic site observations of projects in construction. Watching a building take shape that I had a hand in designing was always immensely rewarding. In this post, I'll consider what happens to that obligation now that AI tools capable of near-continuous site observation exist. Neither AIA Document A201-2017 nor Oregon  Administrative Rule 806-010-0050 says a word about artificial intelligence, which is understandable; neither document has been substantially rewritten lately. I suspect the "periodic" standard was always a practical accommodation to what a human being could manage, rather than a considered position about what is ideal.

I wrote a piece three years ago about OpenSpace AI, another on the architect’s standard of care, and a third, more speculative post on AI’s broader impact on the profession. I believed OpenSpace is a game-changer because it let me follow a project’s progress from anywhere while virtually touring it as if I were physically present. In today's world, where accountability is paramount, I said that OpenSpace offered a superior solution for documentation of construction progress.

A201-2017 draws a deliberate line between the architect's responsibility for "periodic" presence on the jobsite and the "exhaustive or continuous" inspection it explicitly disclaims, and between "observing" the work and "inspecting" it. Oregon's rule sets a parallel standard: observation on "a periodic basis as is necessary," without the same explicit contrast, but pointing in the same direction.

We tend to talk about “observe versus inspect” as if it were a principled distinction about the architect's proper role. Design intent belongs to the architect, while means and methods belong to the contractor, and so on. This allocation of risk is real and defensible. But underneath all this lies a basic fact: the standard assumed a person could reasonably meet it by driving to the site with a notepad and a camera, at intervals consistent with the project’s scope and complexity. It never claimed to be ideal. It simply reflected what was possible. What AI is capable of is bringing this into question.

With apps like OpenSpace AI, continuous, computer‑vision‑assisted site observation has moved from novelty to a maturing product category. Buildots, a competing example with a growing market share, works much the same way. A worker walks the site regularly wearing a 360‑degree hard‑hat camera; the app processes the imagery, matches it to the design team’s BIM model and the schedule, and generates reports, trade by trade and element by element, documenting what is built against what was supposed to be. It flags deviations, forecasts delays, and produces documentation that once required many hours of a superintendent’s time. Several large and sophisticated builders, such as Fortis Construction (OpenSpace AI), and Turner Construction and JE Dunn (Buildots), already deploy these tools. These contractors have normalized continuous observation for their own risk posture.

In one of its case studies, Buildots reports helping a Danish contractor avoid a potential six-week delay and daily fines of 0.1% of contract value on an 86,000 SF office building project. The company quotes the project's director on the part that matters here: claims are usually a mess of poor documentation, but with continuous data in hand, "we have real numbers and historical data to oppose or validate those claims.”

I cite this case study because it shows what problems this technology was built to solve, and for whom. It wasn’t built for design intent, and not for the architect’s interpretive judgment, but to help address disputes, delay, and claims exposure for the parties who bear that exposure most directly. The physical limitation that once calibrated the “periodic, not continuous” standard no longer constrains anyone. Instead, the issue has become one of cost and adoption, which raises a different kind of question, one the legal system has dealt with before.

In a 1932 ruling, Judge Learned Hand decided a case every first‑year law student becomes familiar with. It involved two tugboats, a storm, and the loss of coal barges. The tugboat owner hadn’t equipped his boats with radio receivers, which by then were cheap, dependable, and reasonably common, though not yet universal, and so missed a storm warning that would have compelled him to seek shelter. He argued that since most tugboats in the area didn’t carry radios either, he’d met the industry standard. Judge Hand disagreed, in language pertinent to this blog post: “A whole calling may have unduly lagged in the adoption of new and available devices . . . Courts must in the end say what is required.” Industry custom, in other words, offers evidence of due care. It does not define it. A whole profession can fall behind, and a court can say so.

I raise this legal precedent not because I think a court is about to hold that architects must deploy or rely on OpenSpace AI, Buildots, or their technological cousins. I raise it because the logic transfers uncomfortably well. If continuous, affordable observation exists, and the contractor down the hall already runs it for their own reasons, “I visited during framing, per my professional judgment” will sound less convincing in a deposition five years from now than it does today. The baseline everyone is measured against is shifting in ways architects do not control.

The architectural profession's response to this shouldn’t be dread. The architect’s standard of care and contractual language deliberately allocate risk. The contractor controls means and methods because the contractor builds the project; as the architect, I controlled design intent because that’s the professional judgment I was licensed to exercise. Having a capability doesn't create an obligation to use it in some new, more exposed way. A full-time project representative doesn't change that either. More eyes on the job would not have relieved me of exercising, and standing behind, my professional judgment.

The American Institute of Architects released a Position Statement on Artificial Intelligence this past January, and a more substantive nine‑point Guidance for the Responsible Use of AI by Architecture and Design Firms last October. Both documents are genuinely current and thoughtful. They affirm that professional judgment is non‑negotiable. But professional judgment is not the same as professional procedure, and this is where a gap now grows. For you architects out there, I recommend reading both the position statement and the guidance document.

What the AIA hasn’t done yet is address the architect’s site visitation obligations in light of rapidly advancing AI technology. Their table of “emerging AI use opportunities” covers design, visualization, business operations, and research. Under project delivery, they offer nothing more specific than “task automation” and “documentation.” They offer even less regarding construction contract administration, observation duty, and the architect’s standard of care in the field.

Notably, when I looked at how platforms like OpenSpace AI and Buildots organize themselves, their websites sort their audience by role into project teams, executives, owners, and construction managers. The architect doesn’t appear in their taxonomy at all. They built their tools to solve the contractor’s and owner’s problems, not the architect’s. Consequently, these rapidly evolving tools do not address the architect’s statutory duty to observe the work in progress. When the architect is absent from the tools’ mental models, others risk defining the architect’s obligations.

There is an actual opportunity here. It isn't that AI will make architects better observers, though it might, but that the rule governing how this technology intersects with our legal obligations is still being written, largely by people who aren’t architects, for purposes that fall outside the profession's domain. AIA has made a real start, but it hasn’t finished. The gap between the studio and the jobsite, in the guidance as it stands, is precisely the gap architects would be wise to fill for themselves before an insurer’s underwriting requirement or a plaintiff’s expert witness fills it for them. The moment for architects to articulate what AI‑assisted observation should and should not mean is now.

I think about the field reports I used to write, the ones I was taught and later taught others to word carefully: observed, never inspected, never approved, unless I meant it. I wonder what those reports would have looked like with the benefit of a continuous, camera‑fed record sitting alongside them. Almost certainly better, and far harder to argue with. Whether that makes AI tools a new obligation or simply a better mousetrap is a question the architectural profession will spend the next few years answering. If I were a young architect today, I would embrace the new technology before a legal decision tells me I’ve negligently fallen behind the standard of care.