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.
Procore Pays $845 Million to Give Its Agents Eyes on the Job Site | ConstructionMagazine.ai
Procore Pays $845 Million to Give Its Agents Eyes on the Job Site
Procore agreed to buy DroneDeploy on 29 July 2026, the day it reported its first GAAP operating profit. Its agents could read every document on a project. They could not see the work.
Procore announced the purchase of DroneDeploy on 29 July 2026 for about $845 million in cash. The same morning it reported second-quarter results: revenue of $375.2 million, up 16 percent, and GAAP operating income of $4.3 million, a one percent margin that the company called its first GAAP operating profit. The merger agreement was signed two days earlier. Goldman Sachs committed up to $700 million in bridge financing. The deal needs DroneDeploy's stockholders and antitrust clearance, and Procore expects to close by the end of the year. As of this writing it has not.
Hugo Pegley, who runs the Construction Industry AI newsroom, wrote that the deal buys "the one thing its AI agents could not read: the jobsite itself." Six days before the announcement, Procore had packaged those agents into paid tiers: a Starter tier with five agents for search, submittal review, RFIs, daily logs, and contract review, a Pro tier with twenty, and an Enterprise tier with a studio for building custom agents. Every one of them reads documents. Meg Baldini, Procore's vice president of corporate development, told Construction Dive that acquisitions "accelerate our strategy, they're never the strategy itself." DroneDeploy is the second AI deal of the year, after Datagrid in January.
Procore buying DroneDeploy for $845M isn't a drone story — it's a data story. Reality capture only matters when the system of record can act on it, and jobsite AI needs eyes.
Ian Martin, a VDC lead with two decades in reality capture, posted that on 21 August. It is the consensus read of the deal. The argument among the people who follow construction software is about the price.
Capture feeds the record; the record feeds the agents. AI-assisted illustration.
What Procore is buying, beyond the drones
DroneDeploy's name undersells the product it had become. The company started with flight planning and mapping for drones, and over a decade added the ground half of capture: 360-degree walkthrough cameras carried through a building on a schedule, robot-mounted capture on sites too large or too hazardous for a person, and progress tracking that compares each week's images against the model and the schedule. What Procore gets is not a fleet. It is a pipeline that turns a site into dated, located images that software can compare, which is the raw material an agent needs before it can say whether the work on the ground matches the work on the record.
That distinction matters for the price. A drone company is worth what its hardware customers pay. A capture pipeline attached to the system of record is worth what every Procore customer might pay to have the record checked against the site automatically. The analysts who found the price hard to justify were valuing the first business. Procore is paying for the second, and the payback math below only works if the second business exists.
The payback math behind $845 million
DroneDeploy's revenue has not been disclosed. Analysts discussing the deal have put it in the tens of millions, which makes the price a multiple that only a cross-sell can justify. The arithmetic is straightforward once the assumptions are stated. Procore reports roughly 3,000 customers spending more than $100,000 a year. If capture sells mainly to those accounts rather than to small contractors, and if about 30 percent of them buy, then each converted customer has to add about a million dollars of product for the deal to pay back in a year, half a million over two years, and $330,000 over three. Every input except the price is an assumption. The chart shows the outputs for what they are: a thought experiment about what Procore must believe.
Cross-sell per converted enterprise customer needed to pay back the price, in dollars (illustrative, 3,000 accounts, 30 percent conversion)
The assumption that matters most for readers of this magazine is the first one. If the payback runs through enterprise accounts, the product Procore builds will be shaped for them. A mid-size contractor with two drones and a Procore subscription is not the customer the math is written for.
What the drone counts, and who stops reporting it
Reality capture on large jobs is a decade old. Contractors of several hundred people run fleets of twenty or more drones with dozens of licensed pilots, flying sites on a schedule and feeding the images into progress tracking rather than marketing. OpenSpace, a rival capture vendor, said in a June press release that more than 1,000 data center projects have used its platform. The buildings where the models are trained are the same buildings getting the cameras.
What the capture replaces is a person's report. On a site with drones overhead, the trucks arriving at the gate are counted from the air, and the superintendent no longer fills in that line of the daily log by hand. On a site without them, a gate man counts trucks with a clipboard, and the number reaches the office when he does. The difference between those two sites is the difference the deal is priced on. For a mid-size contractor, the question is narrower than the deal: do the photos, the drawings, and the daily log sit in one place a model can read, and who approves what gets written back into the record.
Who writes into the record
The system of record is the asset in this transaction, and the capture company is the input. A model layered on a record that already holds the contracts, the submittals, the RFIs, and the schedule can make inferences that a model layered on a photo cannot. That is the case for Procore buying DroneDeploy rather than the reverse. It is also why the fight over who may write into that record has started before the deal has closed.
Agents already write into Procore, and some of them override people. Trunk Tools, whose agents process submittals for general contractors including Gilbane and Suffolk, has said its submittal agent replaces the specification section a human selected in roughly one case in seven when it judges the human choice wrong. In June, Engineering News-Record reported that Procore had cut off Trunk Tools' access to its API, citing the security of customer data, and then bought Datagrid to build agents of its own. The record's owner decides who writes to it. That is the position DroneDeploy's cameras strengthen.
The Trunk Tools episode is the clearest signal of how that position will be used. An agent vendor that reads and writes through the platform's interface depends on the platform's permission to exist. Procore withdrew that permission, said customer data security was the reason, and then bought a company to build the same category of agent inside its own walls. Whatever the merits of the security argument, the sequence tells a contractor two things. The platform will decide which agents may touch its data, and the platform intends to sell those agents itself. Capture makes the record more valuable and therefore makes the gate more valuable too.
For a general contractor the practical question is contractual. Who owns the images a drone captures over the contractor's site, the model the software builds from them, and the inferences an agent draws from the model? Construction Dive reported in August that platforms have started competing for exactly that data to train their agents, and that the terms sit in subscription agreements most contractors have never read for this purpose. A firm that flies its own drones today and uploads to a platform tomorrow should know before the upload whether it can take the images with it when it leaves.
Three sites with a contractor on the record
The robotics news in the same month came with names attached. Bedrock Robotics said its retrofitted excavators were running with no operator in the cab on paying customers' sites, and its release, distributed as a paid press release, named Sundt on a Nevada water treatment plant, Champion Site Prep on a Texas earthwork job, and Zachry on a 1.2 million cubic yard civil project. Founders elsewhere in the same category describe adoption of autonomous heavy equipment as close to zero outside mining haul trucks. Both statements can be true. Three sites with a contractor's name on the release are a fact. The share of the industry they represent is not yet a number anyone has published, and the press-release percentages that travel with launches are the weakest claims in the story.
What a mid-size contractor should do with this deal
Nothing in the transaction requires a mid-size contractor to act, and the product it produces will not ship before the deal closes. What the deal does is settle a question about where capture is going: into the record, under the platform's control, with agents reading it. A firm can prepare for that without buying anything.
Put the photos, the drawings, and the daily log in one place a model can read, and name the person who approves anything an agent writes back.
Read the data clauses in the platform subscription for image ownership, model training, and portability, and ask the vendor in writing what changes after the acquisition closes.
If the firm already flies drones, keep the raw captures under its own storage as well as the platform's, so the history survives a change of vendor.
Treat every agent write-back, from a submittal section to a schedule update, as a change a person signs, and log who signed it.
The deal was priced for the 3,000 accounts that spend six figures a year with Procore. The rules that keep a smaller firm's record its own cost nothing, and they are the same rules regardless of who owns the cameras.