Kanopi and the engine behind the bid

What Kanopi is and the problem it solves

Kanopi is an AI construction platform that designs permit-ready architectural plans and prices permit-quality bids from the same model [1]. It is the estimating engine of LÏEF Development, built first for LÏEF's own bids and then opened to other builders and developers [5][7]. Its public line is "Design it. Price it." [1]

The problem is a split that most of the industry accepts. Architects draw and do not price. Estimators price and do not design. Between the two sits a takeoff, the work of measuring every wall, slab, opening and finish off the drawings, and that takeoff is slow, repeated and a common place for a bid to go wrong. Kanopi's site puts the gap plainly: the Revit model that comes out of design flows into the estimate directly, with no re-drawing, no second takeoff and no gap between the plans and the price [1].

The second problem is who carries the risk. The site says takeoff software is sold as a licence: the customer still does the takeoff, and the vendor's terms disclaim accuracy and tell the user to have a qualified professional verify the output [1]. Kanopi does the measuring, measures twice, and puts a stated tolerance on the face of the document [1].

Kanopi has two services [1]:

• Kanopi Design. An AI team led by a human orchestrator works with the client to produce a full, permit-ready Revit plan set. A licensed architect reviews and stamps it. The site says the aim is to cut architectural costs by 40 percent [1].
• Kanopi Estimate. An AI team led by a human estimator turns a plan set into a permit-quality bid in minutes, calibrated to the client's subcontractors, overhead and bidding history instead of a generic cost book [1].

How the estimating engine works

The engine reads a plan set, takes off the quantities, prices them from a sourced rate library and assembles a bid. It carries guards that refuse to ship a number the engine cannot stand behind [5]. The steps below are the ones the site and the Common Ground record describe.

1. Intake. The client sends one project with its plan set, a previous bid that shows its format, and how it prices: rates, markup structure, templates. The plans can be a stamped set, a schematic set, or a marketplace print with no dimensions, as PDF, DWG or a model [1][4].

2. Vector takeoff. In September 2026 the front end was rebuilt to measure the drawing's own vector geometry instead of a picture of it. A CAD-generated PDF stores its lines as paths, and measuring those paths at the sheet's scale gives lengths and areas directly, where a picture has to be interpreted pixel by pixel. The Common Ground record puts the scale at about 49,000 paths on a single plan sheet [5].

3. The overlay a person must inspect. No quantity is trusted until a person has looked at the overlay of what the engine measured drawn over the drawing itself. A ratio check sits beside it as a second control [5]. The Common Ground record puts it simply: a human opens the set and reviews it, Kanopi is expected to keep improving, and the check is too important to skip, so a human overlay stays on top and takes about a hundredth of the time [6].

4. The blind second pass. Every takeoff is measured twice. The second measurement is independent and blind: it has no access to the first, and the two are reconciled before anything ships [1][5]. The rule came from a real catch: on takeoffs run for a building-materials manufacturer, the blind second pass found a wall-height error in the engine's own first pass that would have undersized the shell by roughly a third [5][6].

5. Calibration to the client's shop. The site's position is that two estimators can measure the same building identically and land 50 percent apart, because bidding style is a set of conventions: scope tier, who supplies material, where the furniture and equipment boundary sits, the markup stack. So no live bid ships until calibration closes the drift on the client's own history. The client supplies two or three bids it won, Kanopi reproduces one blind, and the two are reconciled line by line on the client's rates, markups and scope boundaries until the number lands inside the stated band for the plan stage. The site adds that a total that lands by compensating errors does not pass [3].

6. Pricing inputs and confidence on every line. The site names a BLS Producer Price Index feed, daily material prices from 1build, and a live tariff watch on imported categories [2]. Each line states where its number came from: quoted by a subcontractor, measured from the plan, assumed from history, or budgeted as an allowance [1][2].

7. Delivery in the client's format. Work comes back line by line, quantity times rate, inside the client's own spreadsheet and template. The site states that a full takeoff is delivered in 24 to 48 hours, and that a first bid comes back within 72 hours of completed intake [1][4]. The Common Ground record says a full quantity takeoff went from about 40 hours to about 40 minutes, with the blind pass and overlay included, at an internal cost under $350 [5].

Blind accuracy: what Kanopi publishes and why

Kanopi publishes results and keeps the method private [5]. The results it publishes are the ones it earns when the engine has not seen the answer.

The 11-bid blind backcast. Before anything went on a public page, Kanopi re-priced 11 complete past bids from the original intake plan set only, without seeing the bid that was sent. The projects ran from accessory dwelling units and custom homes to restaurants, commercial shells, a 179,000 SF office interior, a 23,000 SF youth facility and a 121-unit hotel, at plan stages from concept sketch to permit-issued sets. The safeguards were a logged blind seal, a visual pass on every plan sheet and explicit denominators on every rate [3].

What measurement got right: on the 121-unit hotel, building area landed within 1 percent, 122 individual quantities matched exactly and the window count landed within 4 percent. Wall systems landed within about 2 percent, and vendor quotes were reproduced exactly in every bid that carried them [3].

What the totals did: uncalibrated bottom-up totals, using the standard rate library with no per-client tuning, landed a median 35 percent from the shipped number, best 2.5 percent and worst 152 percent. Four of the eleven fell inside the margin-of-error band declared for their plan stage. Every miss outside the band traced to a pricing convention, and none traced to takeoff quantities [3]. The site calls that split the finding that matters: the remaining error lives in how a particular shop prices.

The held-out figure. The Common Ground record is strict about what counts as accuracy. The only accuracy claim Kanopi makes is the held-out one: priced blind across three single-family bids, the mean error was 18.6 percent and the worst bid was 31.9 percent, as of June 10, 2026, and the figure is re-run as the corpus grows [5]. In-sample figures are not presented as accuracy [5].

The limits it states. The site says n=11 applies to the blind backcast specifically, run on LÏEF's own bid history and scored against bids that themselves carry stated margins of error by plan stage. It claims no universal accuracy percentage and says it will not publish one until calibrated live-bid results exist per client [1][3].

Why it works this way. The audit behind the policy is on record. Jesse Fowler had a frontier model examine the engine for every place it was grading its own homework. It found that 75 of the 167 rates in the catalog had been transcribed from the anchor project's own budget, and sorted all 96 formulas into calibrated, assumed and known wrong (40, 33 and 11). Fowler wired in the missing gates, added 24 hard assertions, and ran a leave-one-out test: each bid removed from the corpus, priced as if unseen and compared with what went out [5]. A calibration pass against a 41-bid corpus of LÏEF's own submitted bids then dispositioned 54 formulas. Of those, 36 could be calibrated and 18 could not, and the 18 carry their debt annotated in the engine [5].

The first real mistake became a permanent gate the same way. A May 2026 quick-service restaurant bid shipped with its markup stack compounding on itself. It was caught, corrected in the second version, and the check now fails the build on its own [5].

Kanopi Design and the Skill Engine

The design side runs on what the site calls the Skill Engine, a six-step loop [2]. Kanopi captures a licensed architect's working session at the operation level, every Revit command and correction, through a 138-tool programmatic control surface. The sessions are encoded as versioned, executable skills with preconditions, tolerances and the code sections they satisfy. Each skill is verified against deterministic geometry checks, egress and clearance math, rated-assembly rules and building-code gates, and an agent must produce the same plan twice from the same inputs before it is trusted once. Skills compose, from wall placement to unit layout to a floor stack to a permit set, and each new project regression-tests the whole library. A licensed architect stamps every set, and each redline is corrected at the skill level. The last step is to run it on Kanopi's own buildings first.

The test case is the LÏEF conversion of a 1971 concrete shell in Midtown Phoenix into 22 condominiums. The structural model has seven levels, 13 grids and 30 columns. The agents caught a north-south bay that was off by 9.5 feet against the field measurement, and an upper-floor setback the first pass had wrong, with no human flag in either case [3]. The site projects about $250,000 in savings against commissioning an outside architect [3]. The Common Ground record adds that a Revit skills library with a human in the loop turned a stalled as-built into a design-development set in a six-week agentic build, cutting architectural production time 65 percent against a 50 percent target [5].

History and milestones

• 2004 to 2023. Tellus Design + Build, Jesse Fowler's prior firm, ran with architecture, estimating and construction under one roof, completing more than 400 projects and over $300M of built work [7]. The site reports 400+ bids and projects and $300M+ of built work behind its logic [1].
• About 2014. At Tellus, Fowler encoded the per-trade metrics subcontractors price from into an in-house estimating system, so the firm could stop waiting three weeks on trade bids before telling a client what a project would cost. A detailed estimate took two days after that [5].
• April 2026. The pipeline from plans to a priced bid is built. The first autonomous run priced a 6,919 SF premium spec home in 3 minutes 31 seconds [5].
• May 2026. The restaurant markup error, the engine's first permanent gate [5].
• June 10, 2026. The held-out accuracy figure is set: mean error 18.6 percent across three single-family bids, worst bid 31.9 percent [5].
• 2026. The 11-bid blind backcast, run before any public page [3][5]. The estimate that took two days at Tellus now takes about two hours at LÏEF [5].
• September 2026. The vector rebuild. Its first proof caught a five-times overcount on a 36,153 SF campus: 1,834 linear feet of interior partitions, not the 9,470 an earlier desktop estimate had carried. An independent blind pass landed within 2.4 percent [5].
• September 25, 2026. Kanopi is in place with eight builders and developers pricing their own work, and over $600 million in proposals has been bid through it nationally in 2026 [5].

Who Kanopi serves

The site's intake lists the people it expects: subcontractors in drywall and framing, concrete and masonry, MEP and other trades; general contractors, residential and commercial; ground-up developers; owner's representatives and project managers [1].

The offer is a free first bid on a real project, in the client's format. The client keeps the bid either way and owes nothing. After that, each bid is priced to the project and the price is stated before work starts [1][4]. Client data is never sold, shared or used to train anyone else's bids [1][4].

The people

The site says Kanopi was built by an operator who builds and ran on his own buildings before it was offered to anyone [3]. The Common Ground record names Jesse Fowler as the principal of LÏEF Development who holds the estimating seat and built and validated the engine [5][7]. The site states that the masters who train the engine are licensed architects three decades into practice and estimators who have bid, built and reconciled $300M+ of completed construction. It does not name them [1][2].

Frequently asked questions

What is Kanopi? Kanopi is an AI platform for construction that designs permit-ready plans and prices permit-quality bids from one model. It is LÏEF Development's estimating engine, offered to other builders and developers [1][5].

How accurate is Kanopi's estimating? Its one accuracy claim is a mean error of 18.6 percent and a worst bid of 31.9 percent, priced blind across three single-family bids as of June 10, 2026 [5]. In the 11-bid blind study, quantities held, and uncalibrated totals landed a median 35 percent from the shipped bid [3]. It claims no universal accuracy percentage [3].

What is a blind takeoff? A blind takeoff is a measurement made without access to the first measurement or to the answer. Kanopi runs a second, independent blind measurement on every takeoff and reconciles the two before anything ships [3][5].

Does Kanopi replace the estimator? No. A human estimator leads the AI team on every engagement, a person inspects the overlay before a quantity is trusted, and an architect stamps every Kanopi Design set [1][2][5].

What do I need to send for the free first bid? One project with its plan set, a previous bid that shows your format, and how you price. Rough is fine. The first bid is free, with no card and no contract, and it comes back within 72 hours of completed intake [4].

How is Kanopi different from takeoff software? Takeoff software gives your estimator a tool, and the estimator does the takeoff. Kanopi does the measuring, measures twice, returns it in your own spreadsheet and puts a stated tolerance on every deliverable [1].

Where do Kanopi's prices come from? They are calibrated to the client's own bidding history and subcontractors, with a BLS Producer Price Index feed, daily material prices and a tariff watch on top. Every line says whether it was quoted, measured, assumed or budgeted as an allowance [1][2].

Sources

1. Kanopi, home page. https://kanopibuild.com/
2. Kanopi, Platform: The Skill Engine. https://kanopibuild.com/platform
3. Kanopi, Proof. https://kanopibuild.com/proof
4. Kanopi, Try your first bid free. https://kanopibuild.com/start
5. Common Ground wiki, Kanopi: an estimating engine with an accuracy claim that survives checking. https://wiki.commonground.ventures/work/kanopi-estimating-engine/
6. Common Ground wiki, Kanopi: questions answered. https://wiki.commonground.ventures/work/kanopi-estimating-engine-questions/
7. Common Ground wiki, Lïef Development + Construction. https://wiki.commonground.ventures/lief-development/