An independent magazine about AI at work in construction. Every story starts on a real job site, with the people who used the technology and the results they will answer for.
Suffolk Put a Software Engineer in the Fort Myers Site Trailer | ConstructionMagazine.ai
Suffolk Put a Software Engineer in the Fort Myers Site Trailer
Cory Greenwood joined the $1.1 billion airport concourse job in February to build AI tools with the project team, one of five engineers Suffolk has placed on jobs since January. What he does, what Suffolk claims, and what is measured.
Cory Greenwood's title is Site AI Manager, and his site is the 14-gate concourse Suffolk Construction is building at Southwest Florida International Airport in Fort Myers, a $1.1 billion expansion due in December 2027. He is a software engineer, with a data science background and previous work at Raytheon, NASA, and the Army Corps of Engineers, and since February he has sat with the project team instead of at headquarters. Software engineering on a construction job site is a brand new concept, he told the Business Observer in July, and he is still learning it.
Suffolk calls the program Jobsite of the Future. It started in January, and by July it had five engineers on five jobs: Fort Myers, Cedar Rapids, Las Vegas, New Haven, and San Luis Obispo. Doug Harrison, the company's vice president of corporate operations, has said he does not know of another contractor hiring engineers and sending them to jobsites. This profile uses the Business Observer's July report from the Fort Myers site, a Construction Dive Q&A with Harrison published 1 July, and a Construction Dive roundup from 12 August. Greenwood has not spoken to this magazine.
Where the program came from
Harrison started at Suffolk as a project engineer in 2018 and now runs the part of corporate operations that decides what technology the company's builders use. His description of the industry to Construction Dive is that it is fragmented and disjointed, and that AI is the first thing that has brought the pieces together. Jobsite of the Future is his answer to a question he puts plainly: the company hands its project teams the standard model licenses everyone in the industry now has, and the question is how to get day-to-day efficiency from them and train the people on the project to use them for document ingestion, change-order comparisons, and RFI creation.
The licenses are the first front. The second, which Harrison says is where he expects transformational change, is the tools and applications built to go hand in hand with the project teams. That is Greenwood's job. The Business Observer's account puts the engineers' work in three areas, design coordination, schedule, and process, and describes the model as roughly ten years in the making, going back to a decision by Fish to invest in data when the industry's productivity numbers started to fall.
Five engineers sent from headquarters to five jobs since January. The one in the trailer is the program. AI-assisted illustration.
What an engineer does in a site trailer
Greenwood's own description of the job is a sequence. He and the project managers sat down and looked at the processes that could be sped up with AI, then worked out how to have software do the early legwork so the team's time went to the parts of the project that need a person. Harrison lists the administrative work that goes through the engineers: schedule updates, requisition reviews, submittal coordination, and shop drawing reviews, produced in minutes from Suffolk's historical project data. The proximity is the point. An engineer who is on the job sees the inefficiency first hand and can build against it the same week.
The specific field tools Suffolk names for the program are five. Voice-driven scheduling that takes a multi-day schedule update down to hours. Design review that finds drawing conflicts and coordination gaps before construction. Procurement and delivery tracking for lead times and supply risk. Computer vision and site intelligence that cut the time spent documenting and recalling conditions. And operational playbooks that answer process questions from the record. None of the five comes with a per-job measurement, and the Business Observer did not print one. The absence is consistent across both interviews: Suffolk describes what the tools do and what the program cost, and gives program-level percentages, but has not yet put a single tool on a single job with hours before and after.
How Harrison describes the two halves of the program (Construction Dive, 1 July)
1Every project team gets standard model licenses and training to use them on change-order comparisons and RFI drafting
→
2An AI engineer is paired with a team in one sector, because mission critical, healthcare, and higher education deliver differently
→
3The engineer builds tools and applications with the team, on Suffolk's own data
→
4Field users test each tool and sign off on usability before it becomes a standard
→
5Data from the standard tools flows up to regional dashboards and the executives who run the portfolio
The rule that comes before the tool
Harrison's Construction Dive interview is mostly about the rule that governs what Greenwood can build on. Consistent technology has to be deployed first to capture the data, and before a tool becomes required, the company has to know what data it will pull and whether that data is a source of truth for managing projects regionally. The same data goes up to the dashboards that Suffolk's chief operating officers, general managers, and presidents use. Data collection and technology deployment, in his phrase, are thought about in the same light.
On build versus buy, Harrison is explicit that Suffolk is a construction company, not a software company. If a capability is at the front of a platform the company already uses, or something in the market can be tested, Suffolk will not build it unless it is bespoke to the way Suffolk works. Its venture arm, Suffolk Technologies, and its BOOST program give it a view of the market it uses to decide. The site engineers build the bespoke layer. The licenses everyone gets are Claude and ChatGPT.
The buying process is a field test before it is a contract. Harrison's example is procurement and delivery, where Suffolk was, in July, looking for a platform to standardize on. The platforms all say they embed AI, he says, but some fit the way Suffolk operates and some do not. Field users have to try the tool, get the results they want, and sign off on usability before it becomes a standard, and standardizing is then a training problem: the right, repeated training for each platform and each person. The engineer on site is there for the tools that pass that test and for the bespoke gap the market does not fill.
The numbers Suffolk gives, and whose they are
Three figures travel with the program, all from Harrison. Since Jobsite of the Future began, Suffolk's safety performance metrics have been cut in half. Schedule variance has improved to more than 80 percent. Margins and cost overruns have improved for the company and its clients. He also says Suffolk, which did $7.36 billion in 2024 revenue by Forbes' count, is on track to be a $10 billion company next year, and attributes that to the data journey. None of the three is defined on the page, none is audited, and the program was six months old when he gave them.
Suffolk's stated investment behind the program, millions of dollars (company figures, Business Observer, 15 July)
The investment behind the program is clearer than its results. Suffolk's chairman, John Fish, put more than $100 million into data and technology about a decade ago, and the company has put roughly $40 million into a Boston building called 100 Mag that houses an AI studio team building and scaling the tools the site engineers use. Fish's statement to the Business Observer calls Jobsite of the Future the company's boldest investment yet.
What the Fort Myers job is
The concourse itself is the largest construction project in the airport's history: a 14-gate Concourse E, a centralized 16-lane security checkpoint, and rebuilt ticketing, baggage, and concessions. Suffolk has been on it for about a year, with tower cranes on the airfield edge and rebar columns up. It is an aviation job, which matters for the program's design: Harrison pairs each engineer with a sector because mission critical, healthcare, and higher education deliver differently, and an airport is its own sector.
Harrison's line on the program, repeated in both interviews, is that Suffolk is not building AI to replace people. His argument is about stretch: growth is fast, people are carrying more responsibilities, and the keystroking tasks are what the engineers are there to take off them so that expertise goes to risk. For Greenwood, the measure of whether that works will show up in the same place it shows up for every superintendent on the job. The concourse is due in December 2027.
What to ask Suffolk, and any contractor copying it
Ask which of the five named tools is running on the Fort Myers job today and which are still in the studio. Ask for the multi-day schedule update that voice scheduling now does in hours, with the before and after in hours on one job. Ask how safety performance metrics are defined, because a metric cut in half is only a result if the denominator is known. Ask what the engineer built that the project team asked for, and what the team declined. And ask how the five engineers share code, because five bespoke toolsets on five sectors is a maintenance bill that arrives in year two.
The model has one thing most contractor AI programs do not, which is a person whose whole job is to sit in the trailer and notice. Whether that person produces a number the industry can use is the question the program has not answered yet, and Fort Myers, with a December 2027 finish and a year already built, is the job where it can.
What a reader should watch for is a number with the job's name on it. Five engineers on five sites since January is enough time for one of them to publish an hours-before-and-after on a single workflow. When Suffolk does, the program stops being an investment story and becomes a field one.