Procore (NYSE: PCOR) opened its Groundbreak 2026 conference in Orlando on 21 October with three AI announcements in one morning. The one aimed at the people actually holding tools is Procore AI Go — a voice-first mobile app that answers spoken questions from a crew’s own project data, now in open beta for customers in North America.
The other two are platform plumbing: a next phase with NVIDIA putting digital-twin simulation into Procore’s Model Manager, and a collaboration with IBM to connect Procore Asset Management to IBM Maximo. Both sit earlier on the availability curve than a collaboration announcement usually implies: the NVIDIA integration is in closed beta, and the IBM work is a stated intention rather than a shipped connection.
Taken together, the three releases describe a company that has decided the hard part of construction AI right now is not reasoning. It is the interface.
What AI Go Does, Precisely
The release is specific about scope, which makes it easy to assess. Crews can ask questions about “project drawings, specifications, RFIs, submittals, daily logs, observations, and inspections” and receive spoken answers based on their information in Procore. There are push-to-talk controls, which the release explains in terms of the environment rather than the feature list: built for noisy jobsites, so that “surrounding conversations or jobsite noise” don’t trigger unintended interactions. And there is conversation history, with session history retained on the device.
That is a read-only product. The release does not use the phrase, but its own roadmap section says so, under the heading “What’s Next for Procore AI Go”: Procore “is developing additional capabilities that will allow field teams to take action directly from their phones.” The listed future capabilities — creating daily logs and observations, drafting RFIs, translating information from images and speech — are the things a foreman would actually want, and none of them are in today’s beta. The Daily Log Agent and Site Safety Agent are also named as future additions to the app, not current ones.
This matters because of how the product is being positioned. Procore’s framing is its “system of intelligence”: helping teams “see what’s happening in the field, understand it in context, and increasingly take action.” That adverb — “increasingly” — is the one word in the sentence that describes where the product actually is today.
Geoff Lewis, SVP of product at Procore, put the case for shipping the asking layer first in terms of walking distance: “Construction crews shouldn’t have to walk back to the trailer to find the information they need. We’re bringing Procore AI directly to the people doing the work, giving them a faster way to get answers right where they are, in a hands free way.”
The customer voice is Andrew Baker, whom the release identifies as chief information officer at Sequoyah Electric, the Washington State electrical contractor. His framing is the one that explains the product choice: “Using Procore AI Go is like calling a teammate who knows the project inside and out… Our people have always picked up the phone to talk through issues with their team, so it feels familiar from day one.”
That is a real argument about adoption rather than a feature claim, and it is the part of this launch we would weigh most heavily: a voice interface asks less of a crew than another screen does. On a jobsite where tools turn over about as fast as the people using them, that is worth something — though whether it holds up is what the Moss study below is for.
The Meta Glasses Are a Closed Beta, Not a Product
The announcement’s most quotable element is the AI glasses. Procore is partnering with Meta to bring AI Go to AI glasses, letting crews “capture first-person photos and access information about those images” by saying “Hey Procore” without reaching for a phone.
The important words are in the availability section: “The integration with Meta’s AI glasses is in closed beta.” ENR’s Jeff Yoders, reporting from the Groundbreak stage the same day, describes the glasses integration as envisioned rather than launched. The release itself says the collaboration “will explore” how glasses could extend the experience, and names no hardware model and no general-availability date; ENR’s report of the keynote adds neither.
Procore is also running a measurement exercise alongside it: a study with Moss Construction, which the release says was independently designed by SBK Technologies, evaluating how AI agents, AI Go and AI-enabled wearables affect administrative workload. Commissioning a study of your own productivity claim is a better instinct than asserting it, and the result — whenever it lands — will be more useful than the launch.
For what the launch claim rests on now: the release cites a joint Procore and Dodge Construction report finding contractors “can recover two to six hours per person every week through deeper technology adoption.” That is a technology-adoption figure, not an AI Go figure, and the release does not present it as one.
Behind the Voice App, the Agents Got Boring — Deliberately
The more revealing Groundbreak news is in ENR’s report rather than the press releases, because it is about submittals.
Per ENR, Lewis unveiled a submittal agent — which ENR describes as similar to an existing DataGrid agent that can update a submittal in every place it is referenced across a set of construction drawings — and reports that the submittal agent is live now. ENR also reports Lewis showing an agent that updates a project’s procurement log, and says a procurement log exists in Procore as a closed beta test. But the item Lewis apparently made the most of was not an agent at all: a bulk submittal workflow that lets a user update up to 500 submittals at once instead of one by one, with each one checked. ENR reports Lewis calling bulk submittals the third most-requested feature from Procore users, and that McCownGordon Construction worked with Procore on it, per Chloe Huxal, a senior construction technologist there.
ENR quotes Lewis with the line that frames the whole event: “We don’t earn the right to talk about digital coworkers, until we’ve fixed the basics.”
That is a notable thing for a platform vendor to say out loud at its own AI conference, and it is consistent with what shipped. The items ENR reports as live are the submittal agent and the bulk editor; AI Go is in open beta and read-only. The agentic items — procurement logs, action-taking in the field — are in closed beta or on the roadmap. Procore acquired DataGrid, whose agent ENR uses as the comparison above, in January 2026.
This is a different posture from the one we assessed in June, when we went through what Procore’s Copilot, agents and $300M R&D budget had actually shipped against the ambition of the announcement. The ambition has not shrunk. The sequencing has got more conservative, and the marketing has started admitting it.
Where AI Go Sits Commercially
AI Go is not a separate purchase. The release says it is “available today in open beta for customers in North America included across Procore’s Digital Coworker Starter, Pro, and Enterprise packages” — the packaging Procore introduced on 23 July 2026, which runs from a five-agent Starter pack through a 20-agent Pro tier to an Enterprise tier with Agent Studio for building custom agents.
What none of this discloses is price. Procore’s pricing page publishes no dollar figures at all; it says the company charges “an upfront annual fee by product and based upon your Annual Construction Volume (ACV).” So a contractor reading today’s announcement can learn that AI Go is included in a package, and cannot learn what the package costs without a conversation. That is Procore’s long-standing commercial model rather than anything new to this launch, but it is the gap between “included” and “free” and it is worth naming.
The NVIDIA and IBM Announcements Are Earlier Than They Read
Both of the morning’s platform announcements carry the same structure: real engineering, early availability.
The NVIDIA work puts Procore’s Model Manager and a new Realism Mode, powered by NVIDIA Omniverse libraries, in front of teams on complex projects — to simulate extreme weather, sun exposure and crew and equipment movement before construction begins, and to trace how one delayed delivery ripples through a schedule. It is “currently in closed beta.” Steve Davis, Procore’s president of product and technology, framed it as building digitally first; Vladimir Troy, NVIDIA’s vice president of AI infrastructure, framed it around AI factories: “Building an AI factory requires thousands of decisions across power, cooling, compute, networking and physical infrastructure.” The joint customer named is Switch, the data-centre developer, with Corine Winfield, VP data operations, quoted on 3D visualisation. The release’s market framing — companies investing nearly $7 trillion in data-centre infrastructure by 2030 — is sourced in the release to McKinsey.
The IBM work is earlier still, and the release’s own verbs say so: Procore Asset Management and IBM Maximo “can give” operations teams reliable asset data; operations teams “may start with” access to trusted asset information; the capabilities “would establish” a more connected digital thread. The release closes with an explicit disclaimer that statements about future availability “are not commitments to deliver any feature or functionality.” Procore Asset Management itself is available today; the Maximo connection is a stated intention. Kendra DeKeyrel, IBM Software’s vice president for asset lifecycle management, named the useful target: closing the gap between construction data and asset management so owners are “laying the groundwork for AI, condition-based maintenance, and more proactive asset management.”
The handover problem it describes is genuinely the right one. As the IBM release puts it, operations teams at handover rely on “spreadsheets, emailed files, and separately stored documentation that must be reconstructed or re-entered before it can be used.” That is the same structural gap that CivilGrid’s $26 million round and Trendspek’s bet that the record has to outlive the inspection both address from other directions.
What To Watch
Procore’s boilerplate says more than three million projects have run on the platform across 150-plus countries. Distribution on that scale is why a voice interface from Procore matters more than a better voice interface from a startup — and also why shipping it read-only is defensible. A wrong answer to “what does the spec say about this slab” is recoverable. A wrong daily log, auto-filed by an agent, is a record.
So the question the next release has to answer is not whether the voice works. It is whether the action-taking arrives with the same caution the submittal work got. Lewis’s “fixed the basics” line is the standard Procore has now set for itself in public, and bulk-editing 500 submittals is a better demonstration of it than any of the glasses footage. That assessment is ours, not Procore’s.
Of the three things announced this morning, one is in closed beta, one is a stated intention, and the one that actually shipped cannot yet write anything down. On a platform this large, that is the right order to do it in.
