Every estimator has lived this night. It is 11 p.m., the bid closes at 2 p.m. tomorrow, and somewhere on page 212 of a 400 page spec book is a code reference implying that somebody, maybe you, owns a gas shutoff valve nobody drew. You find it, you price it, you move on to the next 300 pages hunting for whatever else is hiding. Six months later a sub calls to say the flashing detail on the parapet was never in anyone's scope, and everyone on the call quietly checks whether it is coming out of contingency or out of them.

That specific flavor of dread, the scope gap that surfaces at the worst possible hour, is exactly the job a Toronto startup just built an AI agent to do, and it launched with numbers attached instead of just a demo video.

Provision, a preconstruction AI company founded in 2022 by Luigi La Corte and Brendan Ardagh, brought its Scope Agent to general availability on August 4. The pitch is narrow on purpose: read the full drawing and specification set for a project and generate a complete, trade by trade scope of work with a citation back to the exact sheet or spec section behind each line. Engineering News-Record covered the launch in its August 5 preconstruction coverage, framing it against a problem every estimator already knows by feel: scope of work drafting is the part of bid prep that eats 30 to 40 hours of manual reading per project, and it is also the part that quietly does not get done when the bid calendar is stacked. Provision says Scope Agent compresses that work to under an hour.

What made the launch notable is that Provision did not just claim accuracy, it published a head to head benchmark against Claude on plumbing scope extraction. Scope Agent pulled 145 scope items at 91.7 percent verified accuracy, against 96 items and 72.9 percent for Claude, with 18 of Claude's 26 flagged errors traced to invented drawing marks, model numbers, or schedule entries that simply were not in the source documents. Provision also reports a broader 97.01 percent match accuracy on scope item extraction and 96.02 percent on subcontractor trade classification, against a 91.3 percent baseline it attributes to professional human estimators. These are the vendor's own published figures, not an independent audit, but the company was specific about methodology rather than waving a single blended number. CEO Luigi La Corte put the stakes plainly in the launch materials: "A 97% accurate scope sheet with full traceability is a document you can put in front of a subcontractor. A 73% accurate scope sheet with fabricated model numbers is a liability."

Two customer stories anchor the claim in real jobs rather than a lab test. ProWest Constructors says the tool caught sheet metal scope items its own senior estimator had missed on a Santa Barbara county project, and that it has cut the time from contract award to fully issued subcontracts roughly in half since adopting it. EllisDon, one of Canada's largest general contractors, used Scope Agent output to issue a formal addendum on the New Brunswick Museum revitalization project, which Provision believes is the first documented case of AI generated scope flowing directly into a live bid package at a top tier GC.

Provision is not a brand new name in this corner of preconstruction. It raised a 7 million dollar seed round led by Cercano Management, with Y Combinator and One Way Ventures returning, bringing its total funding to 8.7 million dollars, and counts Acciona, Bird Construction, Colas, EllisDon, and PCL Construction among its customers, on top of reviewing more than 100 billion dollars worth of projects on its platform to date. Scope Agent is the company betting that traceability, not raw fluency, is what actually gets AI output trusted on a document that a sub is going to sign a contract against.

For a company built around agentic AI on the other end of the bid, taking a completed scope and turning it into a winning response, this is an honest adjacent problem worth watching closely. The two jobs feed each other: a scope of work an estimator can actually trust is what a response agent has something solid to build on. If the industry's next argument over AI in preconstruction is going to be about whose output you can put in front of a subcontractor without checking every line, traceability just became the metric to watch, not fluency.

Sources