2026-10-23

Issue #8 (Friday, Oct 23, 2026)

🎧 Listen to today's episode · 6 min

Get this in your inbox every morning.

Good morning, it's Friday, October 23, and this is The Daily Punchlist. Procore's Groundbreak in Orlando kept the AI news coming, and the takeaway is that the AI is getting specialists. One agent for lien waivers, one for the generator rental, one that knows how your company writes a sub contract. Get the coffee before somebody asks for a three-week lookahead. Here's the punch list.

Today's theme is AI that knows your job, not just the internet. Procore opened an Agent Marketplace with 15 partner-built agents at launch and put a number on its Magpie 3.0 reasoning engine. McCarthy says it trained a model on more than 100,000 of its own RFIs to predict cost and schedule impacts, Newforma bought Toronto AI startup Lumi AI, and EarthCam pushed its camera detections deeper into Procore punch lists and inspections. Out on the solar fields, Gritt Robotics says it's lined up six million panels with three top-10 EPCs. A new survey says contractors are reshaping teams around AI more than they're cutting them, and DPR has advice for hospitals planning buildings around AI. Plus, Coast raised $35M for a fuel card and payments platform built for the trades.

In today's issue

1. Procore opens an Agent Marketplace with 15 partner-built AI agents

Procore announced an AI Agent Marketplace, a new section of the Procore Marketplace where partner-built AI agents work alongside Procore's own. Fifteen agents are available at launch for customers on the Digital Coworker Pro and Enterprise packages, and using them draws on Procore AI credits. The lineup covers a lot of the job: Track3D, Evercam and TrueLook for reality capture, BiltOn and Breadcrumb on the safety side, Planera for schedule risk, Lumber for certified payroll, Bespoke Metrics' COMPASS for subcontractor prequalification, Egnyte cross-checking lien waivers, submittals and closeout drawings against commitments to confirm a trade partner is cleared for payment, Billy for certificates of insurance, Miter for labor job costing, Arcol for BIM quantities, Kaya AI for supply chain, Provizual for municipal inspections and Wipfli for federal tax advisory. Each agent connects per user, so it only sees what that user can see in Procore, and an admin approves it and picks the projects it can touch. On the main stage, Procore showed a United Rentals example: ask Procore AI for a generator, and the agent checks scope and specs against the schedule, surfaces nearby available options and hands the decision back to the person. Procore also said its Magpie 3.0 reasoning engine scored 78% task success on Procore's own benchmark of more than 200 real-world tasks covering RFIs, submittals, drawings and specs, schedule and cost, 18 points ahead of the closest frontier model it tested. Bernards project manager Joseph Kim used Procore AI to review an emergency generator submittal, and it flagged a requirement in a separate electrical spec section: the automatic transfer switch had to come from the same manufacturer as the generator. "It caught something from a different spec. I probably would have missed that."

The fine print: the benchmark is Procore's own, and Procore says full results will be published "in the near future." Magpie 3.0 is still in open beta through Procore Explore.

Why it matters: Nobody's going to buy one AI that does everything well. A marketplace where the payroll specialist, the camera company and the lien waiver checker each bring their own agent, inside the permissions you already set, is how this stuff gets onto real projects. Watch the credit meter, though. Every agent on the job eats from the same budget. Like any good sub, check its paperwork before it starts.

Source: https://www.procore.com/blog/introducing-ai-agent-marketplace (Procore, Oct 22) · https://www.procore.com/blog/ai-that-understands-the-job-not-just-the-document (Procore, Oct 22)

2. McCarthy trains an AI model on more than 100,000 of its own RFIs

In a Palantir announcement about customers building their own AI, McCarthy Building Companies said it fine-tuned and benchmarked an NVIDIA Nemotron model trained on more than 100,000 of its production RFIs to predict cost and schedule impacts. It has since fine-tuned a small Nemotron model in Palantir Foundry to classify RFIs, so project teams can spot emerging risks and prioritize issues before they escalate. It also built an AI-assisted workflow, with a post-trained Nemotron model, for the slow job of turning high-level scopes of work into the Exhibit 1 contract language that spells out what each subcontractor owes. Rico Bertucci, McCarthy's VP of strategic initiatives, said the model "outperforms the frontier at a lower cost" and added: "No one else can pick it up, because it is tuned to how we build."

The fine print: that McCarthy's models beat frontier models at lower cost, and that the subcontract workflow saves 30,000 hours a year, are McCarthy's and Palantir's claims. The release doesn't show how they were measured.

Why it matters: Every big GC is sitting on years of RFIs that say exactly where its drawings go wrong and what it costs when they do. McCarthy is betting that the edge is in that history, not in which chatbot you subscribe to. Your RFI log is a training set. Start keeping it like one.

Source: https://www.businesswire.com/news/home/20261021782963/en/Palantir-Accelerates-Delivery-of-Sovereign-AI-to-New-Wave-of-Customers (Business Wire, Oct 21)

3. Newforma buys Lumi AI to answer questions across project data

Newforma, a construction information management software company, acquired Lumi AI, a Toronto startup founded in 2023. Terms weren't disclosed. Lumi's technology will extend Vojo, the assistant Newforma introduced in May, with plans for natural language search and questions that span multiple projects. Newforma says its Konekt dashboards will show which consultant is slowing RFI turnaround, which project's issue pattern matches one that went wrong before, and where a submittal cycle is drifting from the firm's own baseline, with rollout details to come. Newforma says it holds its AI answers to three tests: explainable, verifiable and accurate enough to act on, and Lumi's engine asks clarifying questions when a prompt is vague instead of guessing. Lumi co-founder and CTO Tudor Boiangiu joins as VP of AI engineering and the team stays in Toronto. Newforma CEO Peter Cannone told Construction Dive there were no layoffs at Lumi and its product won't be sold to non-Newforma customers.

Why it matters: "Which consultant is slowing our RFIs?" is a question every PM can answer by gut. Having it on a dashboard with the numbers behind it changes the conversation at the next OAC meeting. Gut feel is free. Receipts win the argument.

Source: https://www.prnewswire.com/news-releases/newforma-acquires-lumi-ai-and-sets-its-standard-for-ai-answers-on-project-data-302914832.html (PR Newswire, Oct 21) · https://www.constructiondive.com/news/newforma-acquires-lumi-ai/833635/ (Construction Dive, Oct 21)

4. EarthCam fills Procore punch lists and inspections with its own jobsite images

EarthCam premiered the tenth generation of its Procore integration at Groundbreak. Teams can ask questions about equipment, deliveries and PPE violations and get answers through Procore Datagrid. Scheduled Milestone Snapshots upload to Procore automatically with a schedule overlay, punch list items and inspections can be populated with EarthCam imagery directly, and observations show up as markers on the camera archive timeline. EarthCam also added new detections for trench safety, housekeeping hazards, delivery verification and dump truck analysis. It says the integration already shares imagery to more than 23 areas of Procore and automates uploads to 10 Daily Log areas, and the two companies have worked together for 23 years.

The fine print: the release doesn't give accuracy figures for the new detections.

Why it matters: A punch item with a timestamped photo already attached is a punch item nobody argues about. Pulling the camera feed straight into the punch list and the inspection saves the walk-and-shoot round. The camera was on site all along. Now it's doing paperwork.

Source: https://www.globenewswire.com/news-release/2026/10/22/3386713/0/en/earthcam-premieres-tenth-generation-procore-integration-with-agentic-ai-and-visual-intelligence-for-construction.html (GlobeNewswire, Oct 22) · https://www.enr.com/articles/64051-earthcam-brings-more-jobsite-visual-ai-into-procore (ENR, Oct 22)

5. Gritt Robotics says it's lined up six million solar panels with three top-10 EPCs

Gritt Robotics, a California company that retrofits commercial construction equipment and robot arms with AI control, told pv magazine USA it's working with three top-10 solar EPCs to install six million solar panels over the next two years, about 2.8 GW. Its systems are already deployed on Burns & McDonnell solar sites. CEO Puneet Puri said "outdoors is hard": trenches are open, people move through, and materials may have snow, rain or dust on them. His pitch is to put a new brain on proven machines: "Caterpillar and Bobcat know how to build good vehicles." Gritt plans to take what it learns on solar fields to data centers and other infrastructure next.

The fine print: the six million panels and three EPC customers are Gritt's own figures, and the EPCs other than Burns & McDonnell weren't named.

Why it matters: Utility-scale solar is the best audition a field robot can get: the same task thousands of times, in real weather, on a real schedule. If it holds up there, data center sites are next on the list. Six million panels is a lot of reps. That's how anybody gets good.

Source: https://pv-magazine-usa.com/2026/10/22/gritt-robotics-brings-ai-assisted-robotics-to-bolster-shrinking-solar-workforce/ (pv magazine USA, Oct 22)

6. Survey: AI is reshaping construction teams more than cutting them

A BRG survey of 109 builder and owner executives involved in AI implementation, reported by Construction Dive, found that among North American respondents, 46% have changed team structures because of AI, 38% have retrained staff, 18% added headcount in some areas and 11% reduced it. More than two-thirds have invested at least $250,000 in AI over two years, and 37% at least $1 million, but only 15% say it's fully integrated. The biggest worries were liability for AI errors (43%) and a lack of talent (40%). Construction also had the lowest reported AI policy coverage of the four sectors BRG has surveyed: 42% for human oversight and accountability, 34% for accuracy and reliability.

Why it matters: The money is going in faster than the rules are getting written. If an AI tool helps price a job or flag a schedule risk, somebody needs to own the check before it goes out the door. Write down who signs off. Then write down who signs off on the sign-off.

Source: https://www.constructiondive.com/news/construction-ai-reshape-teams-replace-jobs/833612/ (Construction Dive, Oct 21)

7. DPR tells hospitals to plan for AI changing operations before buildings

DPR Construction released research on how AI is changing healthcare operations and when that shows up in buildings. The core problem: AI changes fast, while major healthcare facilities take five to 10 years to plan and deliver and stay in service for 30 years or longer. DPR's study finds operations will likely change before buildings do, and space is more likely to be redistributed than cut, with less need for some administrative functions and more for higher-acuity care, virtual operations, automation and technical infrastructure. Power, cooling, connectivity and data readiness may set the pace, so DPR recommends building in infrastructure headroom and keeping capital plans flexible with phased investment and clear decision points.

Why it matters: For anyone building hospitals, the questions are moving earlier: how much power, how much cooling, how many pathways for what's coming next. Spare capacity is cheap on paper and expensive after the drywall is up. Leave room in the conduit, not just in the budget.

Source: https://www.dpr.com/media/press-releases/ai-requires-healthcare-leaders-to-plan-for-multiple-futures-according-to-dpr-construction (DPR Construction, Oct 22)

Trending on X

  1. @LesyaHendrix (Lesya Hendrix, founder and CEO of Arn Autonomy), on why she'd keep a four-meter, one-ton hydraulic humanoid off a rebar mat: 165 likes, 16 reposts, 1 reply, 107,701 views. Her point: a mat is steel bars with open gaps, and weight plus legs is exactly what you don't want there. She's pitching her own small rebar-tying robots, so she has a horse in this race, but anybody who's walked a mat knows she's right about the footing. https://x.com/LesyaHendrix/status/2112868578135314594
  2. @origin_autonomy (Origin), hiring lead AI, robotics and software engineers after announcing $20M in total funding for its interior construction robots: 37 likes, 6 reposts, 2 replies, 1,462 views. Funding first, then the crew. Same as any job. https://x.com/origin_autonomy/status/2113504678973771925

The Daily Punchline

McCarthy trained an AI model on more than 100,000 RFIs. Somewhere, a project engineer read that and said, "Only 100,000?"

Tool of the day: Compass (Bespoke Metrics)

With Bespoke Metrics' COMPASS agent among the first in Procore's new Agent Marketplace, it's today's pick from our directory. COMPASS handles subcontractor prequalification and risk scoring, giving each sub a Q score, and it's Procore's preferred prequal provider. Its marketplace agent brings COMPASS prequalification scores and capacity limits into Procore AI, so estimating teams can check bidder qualification and overall risk before awarding contracts. https://dailypunchlist.com/tools/finance-analytics/compass/ · https://www.bespokemetrics.com

Funding roundup

Tools in this issue

Browse all construction AI tools in the directory.

Get this in your inbox every morning.