Firms in our early-access beta often ask what a flag means before they ask how many checks ran. That pattern shows what AI compliance checking does in practice.
For architects, the useful question is not whether a system can replace a code review. It cannot. The question is whether it can find likely problems while the design is still easy to change. Kestrel Labs checks a building model against code during design, then gives the project team a focused list to investigate.
What the System Reads
AI compliance checking in architecture starts with structured BIM data. It parses model geometry and properties from a Revit model or IFC file. It does not read PDF drawings or interpret design intent, so the quality and completeness of the model matter.
From the uploaded model, the system extracts room areas, door clear widths, wall assembly types, occupant load factors assigned to space types, and whether fire-rated door families appear at openings in rated walls. It maps those values to the rules for the applicable jurisdiction and code edition.
Where It Works Best
For a Denver commercial project using IBC 2021 with Colorado state amendments, the current beta reliably checks occupant loads under IBC 1004, minimum exit counts under IBC 1006, and mixed occupancies and separation requirements under IBC 508. It also checks height and area for the specified construction type under IBC 503 and 506, plus rated assembly penetrations where the wall type is marked fire-rated.
The result is not a pass or fail determination. It is a list of conditions worth examining, with the code section that triggered each flag. In roughly 30 projects processed in our early-access beta from 2023 through 2025, the most useful early-stage results involved mixed-use occupancy identification, egress path counts, and separation requirements. Those are also common sources of first-cycle plan check corrections on commercial projects.
Firms get the most value at two points. At the end of schematic design, checks can expose occupancy classification and mixed-use separation errors while the floor plan can still change without major rework. At the end of design development, they can catch egress path and construction type issues before CD production. In Denver commercial work, second reviews typically take three to six weeks, so timing matters.
Where It Falls Short
Local amendments are the largest gap. Denver uses IBC 2021 as its base code, but city amendments affect areas such as egress illumination and fire suppression in certain occupancy types. The current system covers the primary Colorado state amendments and the core of IBC 2021, not every city-specific modification.
Conditional interpretation is a second gap. Reviewers sometimes rely on informal jurisdictional practice or administrative guidance that is not stated in the code text. Our system can show what a section requires, but it cannot tell a team how the Denver Building and Fire Life Safety Division has applied that section to a Type I podium building with residential space above.
Deferred submittals create a third gap. If fire suppression, structural, or mechanical work comes in under separate deferred submittals, a base permit compliance check does not cover those systems. The output should therefore be treated as a defined part of a review process, not the entire process.
The Responsibility Line
We are deliberate about this: a clean check from our system does not mean a project will pass plan review. Plan reviewers apply judgment to conditions no algorithm fully replicates, including code intent, site-specific conditions, jurisdictional precedent, and administrative discretion. The architect of record holds the license and responsibility for code compliance. Our system lists conditions that warrant examination; the licensed professional decides whether each is a genuine code violation in context and how to respond.
One beta project showed a meaningful reduction in first-cycle comments compared with an earlier project of similar scope from the same firm. At the end of schematic design, team found four occupancy-related issues. None was complicated once identified. The problem on the earlier project was that no one had specifically run those checks before submission.
A Practical Shift in Practice
That example reflects how firms in our beta use the output. A flag starts a conversation between the architect, code consultant, and project team: Is the model accurate? Does the cited rule apply? Does the design need to change, or does the team need to document its reasoning? The software helps put those questions in front of the team earlier.
AI compliance checking works best as a trigger for design conversations, not as an automated clearance mechanism. Firms use the list to focus their own code judgment while the design can still respond.
