TechCrunch reported on 12 October 2026 that Blueprints AI, a Santa Monica company founded in 2025 by Al Dram and Coco Cao and selected for the 2026 Startup Battlefield 200, turns a 2D sketch, a PDF, or, in its second generation, a written description into construction documents: the drawing set a contractor prices and builds from. The set covers architecture, structural, mechanical, electrical, plumbing and interiors. Version 2, generally available in the week of the report, propagates one change across a project and checks designs against building codes.
The pitch and much of the coverage say “permit-ready” and “code-compliant”. The TechCrunch headline promises permit-ready blueprints “in minutes”. Those words describe an outcome: a building department accepting the set. The reporting establishes the product and its founders’ claims. Independent evidence for the outcome does not exist. Blueprints AI (blueprints-ai.com) is a different company from blueprint-ai.tech, which sells AI drawing review.
What plan check actually reviews, and who signs
Plan check is the review a building department, or authority having jurisdiction (AHJ), runs on a permit set against the code edition that jurisdiction has adopted and its local amendments, which often differ from the model code.
The company’s own site, accessed 14 October 2026, is narrower than the coverage. It promises construction documents “built toward code compliance”, lists the International Building Code (IBC), IFC, NFPA, ADA and local zoning rules, and offers a Title 24 sheet, which is California’s energy code. Its FAQ offers to cross-check outputs against a code file the user uploads and recommends finalising zoning decisions during design, so the local check is only as complete as the file supplied.
Accountability stays with the seal. The site says a licensed architect or engineer must review all documents and take professional responsibility before permit submission, with stamping where the jurisdiction requires it. A Lloyd’s Market Association paper on AI in the international errors and omissions market (September 2025) found liability resting with the contracted professional even when AI produces the error, and described generative AI exclusions as a question insurers were still weighing. Engineering News-Record reported on 9 October 2026 that AI exclusions are not about to appear in designer liability policies, and broker Lockton, a year earlier, found that AI-related claims had not yet emerged while underwriters increasingly ask firms how they use AI. The company’s line that most policies cover AI-assisted work is vendor material. Sealing is largely a North American mechanism; UK and EU architects answer to bodies such as ARB and RIBA, and liability follows the professional either way.
The municipal evidence covers pre-screening, not drawing generation
The strongest independent data on AI and permitting covers a different product class. Seattle’s Innovation and Performance department tested CivCheck from March to October 2025 as an applicant-side pre-screening tool for small residential projects, with the analysis running into early 2026. Applications checked before submission still went through normal city review. Completeness checks were 87% accurate and design-compliance checks 92%.
Seattle’s published evaluation summary supplies the limits. Accuracy is not coverage: removing a review cycle requires detecting every correction a staff reviewer would find, and comparing a design to “the thousands of requirements in Seattle’s building codes” involves requirements that are “conceptually complex and likely difficult to automate”. Seattle recommended a production pilot for completeness only. Honolulu has required CivCheck for residential permits since 1 September 2026, and Q1 2026 data showed decision times for smaller projects falling from 73 days to 32.5 days, alongside a September 2026 report of applications up 34% and a backlog.
No independent, non-vendor evidence was found that an AI-generated or AI-assisted drawing set has passed plan check in any named jurisdiction, and this article did not test the software. Research-stage work on automatic code review (arXiv 2510.02634, October 2025) found deterministic check engines outperforming free-form language-model reasoning on one energy-code strand, which speaks to how automated checking may work; it says nothing about approvals.
What $3,200 to $8,500 a month means for a small practice
Generation-1 pricing in the reporting is $3,200 to $8,500 a month depending on the plan, or roughly $38,400 to $102,000 a year before employer overhead. The Bureau of Labor Statistics put the median drafting wage at $65,380 in May 2024 ($64,280 for architectural and civil drafters), so the subscription costs about 0.6 to 1.6 times one median drafter’s annual wage. The public pricing page now shows “Contact us for pricing” and no free trial. No named customers or case studies were found; the site’s “4,000+ companies already signed up” is an unverified count of signups.
The drafting share is the founder’s estimate
Al Dram told TechCrunch that architects spend about 80% of their time drafting and 20% on design, and that he wants to flip those numbers. The site now says 70% and cites an AIA 2023 Firm Survey that could not be verified. A vendor survey of more than 150 firms reports roughly 40/60 on small projects and 20/80 on large ones, which is survey context rather than an established figure for the profession. Training used eight years of approved sets from Dram’s own design-build firm, per the reporting. Approval in one jurisdiction does not establish compliance in another, and structural analysis and electrical panel design remain on the version 3 roadmap.
For a small practice, the practical questions are how many corrections come back on the first review, which licensed professional will review and seal the set, and who answers when the department refuses a corrected set. The site’s answer to the first is “fewer plan-check corrections”, a vendor claim; the other two still rest on the person who signs.
