Hector has built structures for twenty-eight years, and the fundamental compact of his trade — that the drawing you receive in the morning is the drawing you build from — has been broken by AI-assisted generative design tools that revise plans twice daily. He is being asked to manage a crew against a target that moves faster than concrete cures.
Hector has been building for twenty-eight years. He started as a laborer at nineteen, became a journeyman carpenter at twenty-four, worked his way to foreman on mid-rise commercial projects in his early thirties, and has since become the person general contractors call when a project is in trouble — when the schedule has slipped, when the crew is demoralized, when someone needs to read a site and make decisions faster than the project management software can update. He is the kind of professional the industry talks about when it talks about the irreplaceable value of experience. He has been involved in three different conversations this year in which that value was described as "a challenge to quantify."
The current project is a mixed-use building downtown — five floors of retail and residential, an underground parking structure, a curtain-wall facade that the architect has been refining with a generative design tool since the project broke ground. On the Monday I visited, Hector had received his third set of revised facade drawings in eight days.
"The architect sends a PDF," he said. "I look at the PDF and I figure out what changed. I talk to the glass subcontractor. I figure out if what changed can be done given what we've already built. Usually the answer is yes, with some work. But 'some work' is time, and time is money, and the schedule doesn't move for the architect's revisions."
The New System
Last year I wrote about the superintendent on a generative-design job site — a story about a thirty-four-year veteran of the trades watching the design-build relationship reorganized around tools that allowed architects to iterate plans at a speed the construction process cannot match. Hector is a different story in the same chapter. He is a foreman, not a superintendent, and his perspective is more granular: not the strategic picture of the whole project but the daily reality of a crew trying to build something that keeps changing.
AI-assisted generative design tools have become standard in commercial architectural practice over the past three years. The tools allow architects to explore thousands of design variations based on specified constraints — structural requirements, energy performance targets, aesthetic parameters — and to refine designs based on simulation data that previously required weeks of engineering analysis. The productivity gains on the design side are real: design cycles that took months can now take weeks, and the ability to optimize facade performance against energy models has produced genuinely better buildings in measurable ways.
The challenge — the one Hector lives with every day — is that the tools operate on a different time scale than the construction process they are meant to serve. A generative design revision takes minutes to produce. Revising a submitted shop drawing to reflect it takes days. Sourcing and fabricating the revised components takes weeks. And the building, once steel is erected and concrete is poured, cannot simply be re-run. The material reality of construction has a clock that generative design does not recognize.
What the Data Says
The Associated General Contractors' 2026 workforce survey found that foreman and superintendent tenure — historically among the most stable positions in construction, built on experience and relationship — has declined 14 percent since 2020, with higher rates of departure reported on projects using AI-augmented design tools. Exit interviews cited "design instability" and "unmanageable scope changes" as primary reasons in 38 percent of foreman departures from those projects, second only to compensation.
The Dodge Construction Network's analysis of rework costs on commercial projects found an 8 percent increase in rework expense on projects using iterative generative design compared to projects with more stable design development processes. The increase is not attributable to errors in the design itself — the designs produced by generative tools are typically more carefully analyzed than traditional designs. It is attributable to the mismatch between design revision cadence and construction execution cadence: changes that arrive after a phase of work has begun require undoing what was done.
McKinsey's capital projects practice estimates the average cost of a rework event on commercial construction at $47,000, and notes that the frequency of rework events is the primary driver of project cost overruns — more significant than labor productivity, material cost volatility, or weather. A 4.2 times increase in design revision frequency, even if each revision is small, produces a rework environment that erodes the efficiency gains the generative design tools were supposed to enable.
|
"The drawing exists in a computer. The building exists in the world. The world doesn't have a version history." |
What the Worker Says Back
Hector is not opposed to the technology. He has worked on projects where the generative design tools were used well — where the major design decisions were resolved before groundbreaking, and the tool was used during design development rather than during construction. Those projects ran better than average. He has no quarrel with the tools used at the right point in the process.
His quarrel is with what has happened in the handoff — the moment where the architect's fluid, iterative process meets the construction process's need for stability. The tools have made the design side faster and more iterative. Nobody has redesigned the interface between design and construction to accommodate that speed. The shop drawing submission process is the same as it was. The change-order process is the same. The schedule contingency is the same. The foreman standing between the revised drawing and the crew is the same, except now there are 4.2 times as many revised drawings.
"What I need is for someone to tell me at the beginning: here is the design, it is finished, build it," Hector said. "What I keep getting instead is: here is the design as of this morning. We'll let you know if anything changes. And then something always changes."
He has raised this in project meetings. He has put it in writing. The general contractor's project manager has flagged it to the architect. The architect's response, delivered through the PM, was that the generative design process requires flexibility through construction to respond to real-world site conditions. This is not wrong. It is also not what Hector is describing. He is describing revisions that are aesthetic or optimization-driven, not site-condition-responsive. The distinction matters. The project management structure has not developed a mechanism for making it.
The Contradictions
The building industry's technology conversation is dominated by efficiency: faster design, faster permitting, better material estimation, smarter scheduling. The productivity story is compelling. Construction is one of the least productive sectors of the US economy — output per worker has barely grown in real terms since the 1960s — and technology adoption is widely cited as the mechanism for closing that gap.
What the efficiency story tends to elide is where the inefficiency currently lives and who bears the cost of the transition. When design processes become faster and more iterative, the instability that results transfers somewhere. In the current structure of the industry, it transfers to the people closest to the physical work — the foremen and superintendents managing crews against drawings that change on a cadence the crews cannot match. The productivity gain accrues to the design side. The productivity loss accrues to the execution side. Hector is on the execution side.
He is also fifty-three years old and being regularly told, by people who do not work on sites, that experienced foremen like him are the industry's most valuable resource and that retaining them is a priority. Fourteen percent of those foremen have left the industry in five years. The retention conversation has not yet included asking what makes the work harder than it used to be.
A Working Conclusion
The afternoon I was on site, a fourth revision arrived. The change was minor — a door swing adjustment on the third floor that required redirecting a stub-out. Hector read it, called his carpenter lead, and had the adjustment factored into the afternoon's work plan in about forty minutes. He was good at this. The adaptation was fast and professional.
I asked him if it bothered him, the pace of it.
"Not anymore," he said. "I got used to it." He paused. "That might be the problem. I got used to something I shouldn't have had to get used to. Now it's normal. But it wasn't normal. And the guys who couldn't get used to it left, and we called that 'turnover' instead of calling it 'a sign the work had become worse.'"
The drawing for the revised door swing was on the clipboard. The afternoon crew was setting forms on the floor above. The building was going up, one revision at a time.
Figure 9. Stat callout card: '4.2x increase in design revision frequency on AI-augmented projects — driving an 8% increase in rework costs. The efficiency gains are real on the design side. The costs land on…
REFERENCES
1. AGC Workforce Survey: Technology and Field Labor 2026. Associated General Contractors of America (2026).
https://www.agc.org/learn/construction-data
2. Construction Rework Cost Study 2026. Dodge Construction Network (2026).
https://www.construction.com/toolkit/reports/
3. Capital Projects: Closing the Productivity Gap. McKinsey Global Institute (2025).
https://www.mckinsey.com/mgi/our-research
4. AI in Architecture and Construction: The Design-Build Interface. Brookings Institution (2026).
https://www.brookings.edu/topic/artificial-intelligence/
5. Construction and Extraction Workers: Occupational Outlook Handbook. US Bureau of Labor Statistics (2025).
https://www.bls.gov/ooh/construction-and-extraction/home.htm
6. Generative Design and Construction Productivity: Evidence from Commercial Projects. Journal of Construction Engineering and Management (2025).
7. Worker Voice in Construction Technology Adoption. MIT Work of the Future Initiative (2026).
https://workofthefuture.mit.edu/
Comments (0)
Join the conversation!