Every architect and every GC who has ever chased a permit knows the letter. You submit the set, everyone exhales, and six weeks later a single redlined page comes back in the mail. One line item. Corridor B-3 exceeds the maximum dead-end length by three feet four inches. Nobody on a team of a dozen people caught it, because nobody was measuring corridor lengths, they were measuring deadlines. Now the whole schedule slides another cycle while someone redraws a hallway that was, structurally speaking, fine the entire time.

That gap, between drawings that look done and drawings that are actually code compliant, is exactly what two different corners of the industry are racing to close with AI this year, from opposite ends of the submission process.

The freshest piece of it landed today. OFA Group, the Los Angeles based architecture and AEC technology firm behind the BIM automation platform QikBIM, commercially launched PlanAId, an AI tool that checks a set of drawings against building code before anyone submits them for permit. Users upload vector based PDFs exported straight from AutoCAD, and PlanAId checks them against selected California Building Code provisions covering exit door counts and widths, travel distance limits, dead end corridors, and occupancy based egress capacity, the exact category of finding that shows up in the kind of comment letter described above. The commercial release follows a beta that ran from October 2025, and the initial tier allows up to 300 blueprint pages a month plus an AI chatbot for project specific code questions. "Building code review is essential, but it requires professionals to spend significant time moving between drawings, calculations, and lengthy code requirements," said Larry Wong, OFA Group's founder and chief executive officer, in the announcement. That is a fair description of the job PlanAId is trying to compress from days of manual cross referencing into a pass a design team can run before the set ever leaves the office.

The other end of that same submission is also getting an AI layer, and it has been rolling out for longer. CivicPlus, whose permitting platform is used by more than 850 local governments, brought its CodeComply.Ai plan review tool into general use earlier this year, checking submitted plans against IBC, IMC, IPC, IRC, ADA, FHA, UFC, and NFPA provisions and reporting up to a 55 percent reduction in permit review cycles across more than 30 governments now running it. Honolulu is one of the clearest public examples of what that number looks like on the ground. After adopting AI assisted plan review, the city cut average reviewer time per plan from 60 to 90 minutes down to 15 to 20 minutes, a roughly 70 percent drop, according to coverage of the rollout. Singapore has gone further still on the regulatory side: since October 1, 2025, its CORENET X platform has required large new building projects, those at or above 30,000 square meters of gross floor area, to submit through a system that runs automated checks, including clash detection and code classification, concurrently across reviewing agencies rather than one after another, cutting approval time by up to 20 percent compared with the old sequential process.

Put those two halves together and the shape of the trend gets more interesting than either headline alone. For most of this industry's history, code compliance has been checked exactly once, by whoever happens to be reading the set on the government side, weeks or months after the design team has moved on to the next problem. Design teams checked their own work by eye, plan checkers checked it again by eye, and any daylight between the two showed up as a delay charged entirely to the project. What's happening now is that both sides of that handoff are getting an automated first pass before a human ever has to spend an hour finding the one corridor that runs long. That does not eliminate the human plan checker, CivicPlus is explicit that CodeComply.Ai supports reviewers rather than replacing them, but it does mean fewer of those reviewer hours get spent on findings a machine could have flagged in seconds on either side of the submission.

For anyone who prices and schedules work off a permit set, that convergence matters more than the individual product launches do. A code violation caught before submission is a same day fix. The identical violation caught six weeks into review is a schedule hit, a fee, and in the worst case a redesign that ripples into a bid that already assumed a start date. Compliance gaps are, in that sense, close cousins of the scope gaps that turn a clean estimate into a change order down the line: both are problems that were always technically discoverable in the document, just not by the people who had time to look before the deadline. Tools that move that discovery earlier, whether they are reading a floor plan for egress width or a spec section for a missing trade, are chasing the same prize, and 2026 is turning into the year several of them ship at once.

Sources