Lumber, the construction workforce-management platform, announced on 21 October that its Certificate of Insurance (COI) Agent is available to Procore customers. The release, datelined Palo Alto, 21 October, and carried on the trade site Construction Owners Club, which publishes submitted releases, says the agent is “officially available on the Procore Agent Marketplace,” so that “construction teams can now add AI-powered COI review to the platform they already use to run their projects.” The announcement landed the same day the main programme of Procore’s Groundbreak 2026 conference opened in Orlando.
The distribution claim is Lumber’s own, and we are reporting it as that — the company’s own COI Agent page does not mention Procore anywhere. What that page does independently confirm is the product: the same four functions, the same human-in-the-loop limit, and a place in a family the company already markets alongside a Union Agent, a Certified Payroll Reporting Agent, a Funding Agent and a Paper Timecard Agent.
The Workload Being Automated
The release puts two numbers on the problem, both Lumber’s own figures:
- A general contractor “may manage 200 to 300+ active COIs across projects, subcontractors, and vendors.”
- Most teams still do the checking by hand, “spending 10 to 15+ hours a week on review, tracking, and follow-up.”
The second number is the one that explains the product. Ten to fifteen hours a week is the better part of a part-time role, spent entirely on comparing two documents that nobody enjoys reading. And the release is precise about why the comparison is hard rather than merely tedious: each certificate “has to be checked against the insurance terms in its subcontract, including coverage types, limits, endorsements, and additional insured requirements,” and “some of those obligations run for years after a project closes out.”
That last clause is the part that is easiest to lose track of. A certificate that expires after practical completion is still the GC’s exposure, and it is the kind of obligation that goes unwatched once the project team has demobilised.
Four Functions, and the One That Matters
The release lists four capabilities:
- Contract review — identifying each subcontract’s insurance requirements, including coverage types, limits, endorsements and additional insured obligations.
- Coverage validation — checking every COI “against the requirements that the subcontract actually sets, not a generic checklist.”
- Compliance insights — flagging missing, expired or insufficient coverage.
- Renewal monitoring — tracking expirations and renewals “across the project lifecycle, including obligations that extend well past completion.”
Function two is the whole argument. A certificate-tracking tool that checks a COI against a house standard tells you whether the sub has insurance. A tool that checks it against that sub’s own subcontract tells you whether the sub has the insurance it promised you — which is the question that matters when a claim arrives. Reading the subcontract is the harder half of the job, and it is the half that document-reading models have only lately been able to attempt; it is the same kind of move we described when Document Crunch extended its contract review into the drawing set.
”It Flags the Gap. Your People Make the Call.”
The release is direct about the limits of the automation, and the sentence doing the work is a quote from Nigel Coelho, Lumber’s general manager for agentic compliance:
“The COI Agent does the reading and the cross-checking, so your team spends its time on the exceptions that actually carry risk. It flags the gap. Your people make the call.”
Under a heading reading “AI-assisted. Human-approved,” the release states that “insurance compliance isn’t something to hand over to AI unquestioningly” and that the agent “never approves or rejects a subcontractor on its own.” It identifies requirements, analyses coverage and highlights exceptions; “then the GC, controller, or compliance team reviews the findings and makes the final decision.” The company’s own product page makes the same commitment.
Shreesha Ramdas, Lumber’s CEO and co-founder, framed the distribution choice this way: “We built Lumber’s AI agents to take on the compliance work that slows contractors down, and the COI Agent is the next step. Bringing it to the Procore Agent Marketplace meets builders where they already work.”
Behind the agent, the release says, sits LumberCare — “Lumber’s support crew of real CPAs and prevailing wage specialists.” A software vendor advertising human accountants behind its AI is a positioning decision as much as a staffing one, and it is consistent with the rest of the release: this is being sold as a reading machine with a review queue, not an approval engine.
Where This Sits in the Stack
Subcontractor risk has drawn a run of announcements this month. On 15 October, Bespoke Metrics said Procore had named COMPASS its preferred subcontractor prequalification solution, where the design question was which data stays outside Procore; Access Construction pushed an agentic layer across estimating, site and ERP the same day. Procore’s own published lineup — twenty named pre-built agents at the time of writing, all of them Procore’s, including a Contract Review Agent, plus an Agent Studio for customers who want to build their own — already covers contract reading in the abstract.
Insurance certificates are a narrower job than that, and narrowness is the point. A COI is a structured-ish document with a short list of fields that matter; a subcontract’s insurance exhibit is prose with numbers in it. Pairing them is a well-shaped task for a language model, the failure mode is visible (a flagged gap that isn’t one), and the consequence of the model being right is a conversation with a broker rather than a change to the building.
What we cannot tell from the release is how well it does the job. Neither the release nor Lumber’s product page carries an accuracy figure, a benchmark, a named customer or a price — and we would rather say so than quote a number nobody published. Those are the questions to put to a GC that has run it across a few hundred certificates, and the ones worth asking before the 10-to-15-hour figure gets quoted back as a saving.
