6 weeks → 45 minutes

A full feasibility study, recompressed

FEASIBILITY · PRACTITIONER TEST

Leni: A Full Feasibility Study in 45 Minutes

Jake Heller and Quinn Edwards handed Leni an unknown Bronx parcel and asked it to do the work a development analyst would spend weeks on. It came back with a 25-page feasibility study, an Excel underwriting model, a construction cost estimate, and a clear go/no-go recommendation—in 45 minutes. Not a polished marketing demo on a known site, but a cold-start analysis on a parcel the system had never seen.

This is the single clearest data point in the issue: the entire upstream analyst function, compressed into one sitting.

AI for CRE Collective · June 8

FEASIBILITY

Feasibly: “Outdated Before the Ink Dries”

CEO Brian Connolly frames the problem the old way: a traditional study takes six weeks and “can be outdated before the ink dries on the binding.” Feasibly’s pitch is verifiable analysis delivered “in days for a fraction of the cost”—and, critically, one that “can easily be updated as market conditions change.”

The value isn’t only speed. It’s that the study stops being a static artifact and becomes a living model.

Propmodo

FEASIBILITY

GroundUp / Aktus: Architect-Reviewed Stacks at 80% Less

GroundUp puts the traditional number at $15K–$200K+ and four-to-twelve weeks. Its platform delivers architect-reviewed feasibility stacks—pro formas, zoning analysis, 3D massing, and site intelligence—in days, at a claimed 80% cost reduction. The “architect-reviewed” qualifier matters: this is positioned as augmented expertise, not unsupervised output.

Note who reviews the stack. The human judgment didn’t disappear—it moved to the end of a much faster pipe.

GroundUp / Aktus

SITE SELECTION

TestFit Site Solver: Parcel to Pro Forma in Minutes

Site Solver folds the data layers a site analyst chases by hand into one generative engine: cut/fill earthwork cost optimization, SSURGO soil data, ESRI flood and wetland data, and zoning rules for FAR, parking, and height. The schemes it generates can be filtered by FAR, yield-on-cost, and parking ratio simultaneously—so a developer interrogates the trade-offs in real time rather than waiting on a revised study.

The constraint set that used to define weeks of back-and-forth becomes a set of sliders.

TestFit Site Solver

MASSING · LIVE SESSIONS

Zenerate: Massing Schemes Inside the Client Meeting

Sami Khoury, AIA, a Principal at AC Martin, describes using it in the room: “In live client development strategy sessions, it helps deliver massing design schemes based on unit mix and parking ratio inputs.” His verdict goes further—“Zenerate is redefining feasibility modeling and setting a new standard for the market.”

When massing happens live, the deliverable stops being a document handed across a gap and becomes a conversation. That is the handoff collapsing in practice.

Zenerate

ADOPTION REALITY

88% Are Piloting. 60% Aren’t Ready.

A JLL survey of 500+ senior decision-makers found 88% running AI pilots—but 60% reporting they are unprepared to implement AI into their actual workflows. The appetite is nearly universal; the operational readiness is not.

Hold this number against the demos above. The capability is racing ahead of the organizations meant to absorb it—a gap this issue returns to in Section 9.

Propmodo / JLL

CASE STUDY · HONOLULU DPP

“Like TurboTax for Permitting”

Honolulu’s Department of Planning and Permitting, working with CivCheck (Clariti), cut permit decision time from 73 days to 32.5 days—a 55% reduction. Review cycles fell from 3.4 to 1.4. Corrections per permit dropped from 23.5 to 7.7, a 67% reduction in errors. Across a five-month pilot spanning 145 applications, residential plan review time fell 70%.

Director Dawn Takeuchi Apuna put it plainly: “It’s kind of like TurboTax for permitting.” The scale matters—Honolulu processes roughly 20,000 permit applications a year, so a 70% cut in review time is not a pilot curiosity, it is a structural change to the city’s throughput.

The TurboTax analogy is exactly right, and exactly the point: the work didn’t get eliminated, it got guided—the applicant is walked toward a clean submission before a human reviewer ever sees it.

Clariti  ·  StateScoop

70%

Less review time

MUNICIPAL DEPLOYMENT

Denver Takes On a Monthslong Backlog

In May 2026, the Denver Permitting Office launched CivCheck specifically to attack a monthslong backlog that has been throttling housing development. Honolulu is no longer an isolated proof of concept—the same engine is now being pointed at one of the more notorious permitting bottlenecks in the mountain West.

Route Fifty

MUNICIPAL DEPLOYMENT

Edmonton: Same-Day Residential Permits

Edmonton’s AI “auto-review” system delivers same-day residential permits against a typical 20-day wait. When the review is automated on intake, the calendar that developers carry in their pro formas—the entitlement carrying cost—effectively collapses to zero for qualifying projects.

What Works Cities · Bloomberg

MUNICIPAL PILOT

Port Orchard, WA: A “TSA Pre-Check” for Permits

In a one-year pilot with PermittableAI, Port Orchard’s system is already catching roughly 85% of the comments city staff were making by hand. The stated goal is a “TSA Pre-Check style permit track”—a fast lane for applicants and projects the system already trusts.

The metaphors keep arriving from adjacent bureaucracies—TurboTax, TSA Pre-Check—because the move is the same: pre-clear the routine so humans can spend attention on the exceptions.

Kitsap Daily News

MUNICIPAL DEPLOYMENT

Manvel, TX: First in Texas

Manvel became the first Texas city to deploy AI permit review, using Blitz AI to automate building and planning approvals. The early-adopter map is no longer coastal-only—the Sun Belt growth markets, where permit volume is exploding, are wiring up too.

SubcUSA

PROGRAMMING · THE MERGE

Programming and Pro Forma, Fused

The clearest signal of the collapse lives here. As noted in Section 2, Zenerate now delivers massing, unit mix, parking, and pro forma simultaneously inside live client sessions. Programming—historically a discrete, billable pre-design phase—is no longer a sequential step that produces a document for the financial model to consume later. The two are computed together, in the same breath.

When the program and the money are solved at once, the question of which discipline “owns” that work—and bills for it—stops having a clean answer.

EARLY MEP · ENGINEERING

AFRY + Endra: AI for Electrical & Fire Safety, Early

On April 23, AFRY—a major Nordic engineering firm—announced a strategic collaboration with the Swedish startup Endra to bring AI into electrical and fire-safety design at the earliest project stages. The framing is deliberately conservative: AI complements rather than replaces engineers, freeing them for analysis and strategic decisions rather than rote coordination.

A large incumbent engineering firm choosing to push AI to the front of its own process—rather than waiting for it to arrive—is a different signal than another startup demo.

EuropaWire · April 23

EARLY MEP · THE STAKES

Why MEP Is the Right Target

MEP accounts for up to 30% of total construction rework—the single fattest target for early-stage intervention. AI coordination tools claim to cut coordination time by 70%+. The economics are obvious: a clash resolved in a model costs a revised line; the same clash resolved in the field costs a crew, a delay, and a change order.

Vavetek

ACADEMIC · MDPI BUILDINGS

Coordinating at LOD 100–150

Yu-Pin Ma, writing in MDPI Buildings on April 29, proposes a serious-gaming framework for early-stage MEP coordination at LOD 100–150—the conceptual, low-detail end of the model. The premise is the whole argument of this section in miniature: conflicts caught at LOD 100–150 are orders of magnitude cheaper to resolve than the same conflicts discovered at IFC.

The research and the firms are converging on the same instinct: push coordination as far upstream as the model will bear, because that is where error is nearly free to fix.

MDPI Buildings · April 29

ADOPTION · BUILTWORLDS 2026

Estimating Software Is Now Table Stakes

BuiltWorlds’ 2026 Preconstruction Benchmarking report finds roughly 90% of precon professionals have implemented estimating software, and 64% now use it on every project—a 42-point year-over-year jump, and a 392% increase from 2022. Estimating is no longer an early-adopter behavior. It is the baseline.

BuiltWorlds

TAKEOFF · PEER-REVIEWED

Togal.AI: 76% Faster, 98% Accurate—On the Right Geometry

A peer-reviewed study out of the University of Kansas found Togal.AI delivered takeoffs 76% faster than traditional software, with up to 98% accuracy on structured project types. The qualifier—“structured project types”—is doing quiet but essential work, as the next card makes painfully clear.

Contractor Guide Pro

THE HONEST NUANCE · CONTROLLED TEST

When Claude Did a Takeoff With No Human Guidance

Luigi La Corte at Provision ran the test the marketing decks avoid. He had Claude generate takeoffs from a 147-page structural set and a 156-page architectural set with no human guidance, then compared the result to a professional consultant’s number.

Where the geometry was bounded and repetitive, it was excellent. Concrete volumes came in within 2%. Steel column counts were perfect. But where the geometry got irregular, it broke down badly: secondary framing was underestimated by 50–96% on six of eight beam types, and the building envelope was systematically underestimated because the model used the plan footprint rather than the true 3D surface area.

The net result was a bid of roughly $443K against the consultant’s ~$907K—a 51% underestimate. A number that confident and that wrong is more dangerous than no number at all.

$443K vs. $907K

Unsupervised AI bid vs. consultant · a 51% miss

CASE OUTCOMES · SUPERVISED USE

What It Looks Like When Humans Stay in the Loop

Seagate Development Group — Saved 40–50 estimator hours a month with Palcode.ai (about three man-months a year) and avoided an estimated $35K–$60K of change-order risk per project cycle.

HJD Capital — Got roughly 400 estimator hours a year back, with a path to scaling that to ~1,000—the equivalent of a senior estimator recovered without a hire.

Provision — GCs are cutting scope-review time by up to 80%, assembling scope packages in under 60 minutes versus 30–40 hours by hand.

The ROI is real where the work is bounded and a human owns the result. The same tool, unsupervised, produced the $443K miss above. Both facts are true at once.

Palcode · Seagate  ·  Boon · HJD  ·  Provision · ROI

THE STAKES

The Cost of a Document Nobody Reconciled

Change orders run 5–15% of project value, with a median approval time of 14.3 days (FMI). Roughly 30% of RFIs trace back to cross-document inconsistencies, at an average of $1,080 per RFI (Navigant). Rework attributable to cross-document issues runs 5–9% of project cost (CII). These are not exotic risks—they are the ambient friction of every project, and they originate in documents that disagree with each other.

US Tech Automations  ·  Helonic

CHANGE ORDER AUTOMATION

14.3 Days to 5.7

Automated change-order workflows have compressed median approval from 14.3 days to 5.7—a 60% reduction (ENR, 2025). Just as consequentially, disputes drop 40–55% when the workflow produces timestamped audit trails. The speed is the headline; the accountability is the lasting value.

DOCUMENT REVIEW · QA

LightTable: 70% of Errors vs. 30% by Hand

LightTable catches 70% of design errors against roughly 30% for manual review, and turns reviews around in 3–5 days instead of 3–6 weeks. It has run across 20M+ square feet and $3.5B in project costs, with clients including Suffolk, Mill Creek, and Swire, on the back of a $22M Series A.

Note the investor: DivcoWest, a developer fund. When the capital backing design-QA tooling comes from the development side, it signals demand for the deliverable from the people who carry the risk—not just from the architects asked to produce it.

RISK INTELLIGENCE · TRIMBLE

Document Crunch: Project-Level Risk, Holistically

On June 9, Document Crunch (Trimble) launched what it calls construction’s first project-level AI Risk Intelligence platform. The framing stat is sobering: disputes “average more than $60 million per dispute in North America.” Its “Project Assist” analyzes a full document set holistically rather than one contract at a time—reading the project the way a dispute eventually will.

Trimble · June 9

CASE STUDY · SFO

Hensel Phelps + Track3D: 3,000 Hours, 3 Reworks Prevented

On the $300M Courtyard 3 Connector at SFO, Hensel Phelps used Track3D to eliminate roughly 3,000 hours of manual coordination, prevent three major reworks, and deliver $342K in verified labor savings—enough that the firm signed an enterprise agreement off the back of it. Thai Nguyen of Diverge frames the why in one line: “You’ve got to do more with less.”

Verified savings on a named project with a signed enterprise deal is about as hard as evidence gets in this space. The reworks that didn’t happen are the ones that count.

Construction Dive

THESIS · a16z

“Designed By Software Built in 1997”

Schmidt, Haber, Goggins, and Elmgren make the provocation in the title—“Every Building You’ve Ever Been In Was Designed By Software Built in 1997.” Their map lays out three lanes, but the near-term wedge they identify is not replacing Revit; it is document review, MEP, and rework. The anchor stat: pre-bid review takes 3–6 weeks, costs $50–100K per project, and still misses roughly 70% of the issues that later become field change orders.

The wedge is precisely the upstream work this whole issue is about—not the drawing, but the checking.

ARGUMENT · KP REDDY

The Project Is the Ecosystem

In “The Project Is the Ecosystem” (June 16), Reddy argues the AI learning loop in AEC must be owned by the project—the temporary coalition that is construction’s true unit of delivery—not the firm and not the model vendor. In “It Was Never About the Technology” (June 4), he sharpens the point to a single line: “Interoperability without accountability is just an expensive file format.”

Better tools won’t fix outcomes unless the commercial structure around them changes. The technology was never the bottleneck—the contract was.

The Project Is the Ecosystem  ·  It Was Never About the Technology

ARGUMENT · COMMON EDGE

The Mispricing Trap

Common Edge asks whether AI will make architectural expertise easier to misprice, and answers with a paradox worth memorizing: “Work becomes easier to misprice precisely when expertise becomes more important.” When AI compresses the visible early-stage production, the market reads the compression as a discount—even though judgment, risk, and accountability are entirely undiminished.

The prescription is concrete: treat early risk analysis as billable, record the reasoning as a deliverable in its own right, and define scope around advisory judgment rather than document output.

Common Edge

REGIONAL · ALABAMA PIPELINE

Four Hyperscale Facilities, One Feasibility Question

The Alabama pipeline alone holds four hyperscale facilities at various stages:

Nebius Oxmoor — Birmingham, Hoar Construction, $40M in permits pulled.

Project Marvel — Bessemer, a reported $14B.

Columbiana — $1.1B, Digi Power X / Cerebras.

Project Red Clay — Lowndes County, $1.5B, Cloverleaf.

The South now holds 50%+ of the national $73.1B data-center pipeline (ConstructConnect). And evaluating sites like these—power, water, soil, zoning, earthwork, yield—is precisely what the feasibility and site-selection AI in Sections 2 and 3 was built to do.

The regional story and the technology story are the same story. The boom is being underwritten, parcel by parcel, with exactly these tools.

Bham Now

COUNTERWEIGHT

Most of the Industry Hasn’t Started

RICS surveyed 2,200+ professionals in 2025: 45% have implemented no AI at all, under 12% report regular use, and fewer than 1% have it deployed organization-wide. BST Global puts AEC firms at widespread adoption at just 1%. Zoom out to the broader economy and it’s no rosier—95% of enterprise AI pilots deliver zero measurable P&L impact, and 85% of failures trace back to poor data quality.

Every case study in this issue is real. So is this: the median firm is nowhere near them. The distance between the two is not a rounding error—it is the actual state of the industry, and the opportunity for anyone willing to close it.

Bridgit  ·  BST Global