BuildOps announced three AI agents on 5 November at Forge 2026, its annual customer conference, from a Los Angeles dateline. The press release calls them the Service Agents and places them inside OpsAI, the platform’s intelligence layer.
What distinguishes the announcement is not the agents themselves — this publication has covered a steady run of agent launches across construction software this autumn. It is that BuildOps attached a number to what they are supposed to achieve, and the number is a staffing ratio.
The Three Agents
Revenue Agent. It reviews technician visit notes alongside quotes, service agreements, equipment records and jobs, and flags five specific failure modes: work “written up but never quoted,” equipment “on a serviced property that isn’t under the agreement,” quotes “drafted but never sent,” quotes “that have gone quiet,” and “approved work with no job on the schedule.” It then drafts the next step — a recommendation, a quote, a follow-up email or a new job. The OpsAI product page puts a threshold on one of those: “Quotes idle 14+ days and assets with no active service agreement get flagged.”
Service Coordinator Agent. It reviews open jobs daily against what the release describes as more than 20 reasons work can stall — a missed visit, parts not received, a not-to-exceed limit exceeded. It sorts jobs by what they need next, drafts messages to whoever has to act, and looks for billable parts, fees and time sitting in visit notes that never reached the job, priced from the contractor’s own price book.
Billing Agent. It applies each customer’s billing rules, reads the contract and builds the invoice. Before it goes out, it flags low markups and negative margins; it also catches completed jobs that were never invoiced, invoices that bounced or went unopened, and jobs reopened after billing.
Two more were previewed rather than announced: a Collections Agent, which the release says gives the team “one queue each morning of past-due invoices, unapplied payments and who to follow up with next,” and AP Automation, which reads vendor bills, matches them to purchase orders and receipts, codes them to jobs and flags exceptions. The release puts both in early 2027.
What Ships When
This is worth stating plainly, because the announcement and the availability are two months apart. The Revenue Agent is in early access now. The full Service Agent Team becomes available to BuildOps customers in January 2027. The release points contractors to Forge or the OpsAI page to sign up.
So one of three agents is in customers’ hands, and the thing being announced is mostly a January product. Trade coverage on 5 November adds that the Collections Agent Team and AP Automation open to a first group of customers in December — a detail this publication has seen in that report rather than in the release, and reports as the magazine’s.
The Ratio Is the Argument
The figure the release builds on comes from BuildOps’ 2026 Commercial Contractor Benchmark Report: shops on BuildOps average about five field technicians for every person in the back office, and the top 20% support nearly seven. Will Lehrmann, the company’s chief product officer, ties the agents directly to closing that gap, and draws the boundary in the same breath: “These agents don’t replace anyone. They take on the work that buries the back office, so people can do the work only they can do. They’re built to help every shop get to seven.”
Steve Chew, co-founder and chief strategy officer, frames the problem the ratio describes: “Our customers have more work than they can staff, and the back office is where that shows up first.”
Analysis. Many of the agent announcements this publication has covered this year have been sold on task-level time savings. A technician-to-admin ratio is a different kind of claim, and a more falsifiable one. It is measurable from the contractor’s own payroll, it does not depend on believing a vendor’s stopwatch, and it fails visibly: a shop that adds agents and still needs a new billing clerk at the same headcount has an answer.
It is also a claim about where the constraint sits. The argument is that these businesses are not short of demand or of field labour relative to demand — they are short of the back-office capacity to convert field work into money. That is a specific and testable diagnosis of the commercial service trades, and it is consistent with what this publication found in ServiceTitan’s own account of AI adoption in the commercial trades, where the gap between adoption and measured return was the recurring theme.
What the release does not provide is the other half of the ratio claim. It gives no figure for how much the ratio has moved at any shop using the agents, and the benchmark figure describes the installed base rather than a before-and-after. The January date means that evidence does not exist yet for the full set.
The Evidence Offered, and Its Shape
The release offers three pieces of evidence. About 16,000 people use OpsAI each month. In early access, one contractor used the Revenue Agent “to act on $2 million worth of quotes that had gone cold in a single 30-day period.” And the benchmark report above.
The release’s wording on the second is “act on.” What that establishes is reach rather than conversion: the release does not say how many of those quotes were subsequently approved, and that is the figure that would show whether the agent surfaced money or surfaced work.
The customer testimony is more specific than most. Collin Drovdahl, operations manager at Jolma Electric, on the Revenue Agent: it “picks up things the guys mentioned in our daily reports that the PM might miss,” and “We’ve been able to realize real revenue with that.” Jonathan Clark, principal at Omnia Mechanical Group, describes the failure the agent is aimed at rather than the fix: “Customers are getting the proposal, reading it and forgetting about it.”
A Person on Every Approval
The governance line is stated twice, in the release and on the product page, and in nearly the same words. The release: each agent flags a problem, drafts a fix and routes it to the person who owns the decision, and “Nothing an agent writes reaches a customer until a person approves it.” The OpsAI page: “OpsAI recommends, your team approves. Nothing moves without a yes.”
That is now the settled pattern rather than a differentiator, and this publication should say so rather than treat it as a finding. Build Paperless put a human gate in front of the money scope when it opened an Australian builder’s job-management software to agents in October. CMiC made permission inheritance the headline feature of NexusAI at the start of this month. NetSuite’s construction agents reach for the ledger with the same review step. When an agent’s output is an invoice to a customer or a journal entry, the vendors in this publication’s coverage have all arrived at the same answer, which is a reasonable sign it is the right one.
What none of these announcements has described is what the review step actually shows. An operations manager approving a drafted invoice with a flagged margin is approving something they did not compose, and the quality of that gate depends entirely on how legible the draft and its evidence are. The release goes further than most on this point: “Every finding shows why the record was flagged and the evidence behind it. Every message sent is stored word for word, and every declined suggestion records a reason.” That is a description of the gate’s materials rather than a measure of whether reviewers read them, and this publication has not tested it.
The Company
Per the release, BuildOps is an AI-native platform for commercial contractors spanning service, projects and financials; it was founded in 2018 and serves more than 1,500 companies across North America, with backing from Founders Fund, N47 and Meritech Capital.
It sells into the commercial specialty trades — mechanical, electrical and plumbing service contractors — which sit at the edge of this publication’s beat. The shape of that work is what makes the back-office argument coherent: service calls are high in volume, repetitive and billed in small units, so every one of them is another chance for billable time to go unrecorded or a quote to go unsent. That is the surface all three agents are pointed at.
