Blueprints AI, a Santa Monica startup, says its software turns a 2D sketch or PDF into a coordinated, code-checked drawing set across architecture, structure, mechanical, electrical, plumbing and interiors, with a licensed professional reviewing and sealing it before permit submission, TechCrunch reported on October 12, 2026. The report includes no accuracy figures, named customers or independent testing, so for now the claim is the company's own.
What does Blueprints AI say its software can do?
If you've ever spent a weekend redlining a set because one wall moved, you'll feel the pull of this pitch. According to TechCrunch, the second version of the software starts from a description or sketch, produces the drawings for every discipline at once, and updates the whole project when you change one element, checking each change against local building codes. Co-founder Al Dram, who opened the design-build firm Drams Design-Build in 2018, says the system was trained on eight years of approved drawing sets from that firm. The first version was priced at $3,200 to $8,500 a month at launch.
Dram estimates that architects spend about 80% of their time drafting and 20% designing, and he says he wants to flip that ratio. The co-founders' framing is that the judgment stays human, because a licensed professional still reviews and seals every set.
What does the launch coverage leave open?
Reading the TechCrunch piece closely, you'll find the claims and very little that tests them. It doesn't say which jurisdictions or code editions the checker covers, how often a generated set passed review, or who has used it on a submitted project. You'd be relying on training data from one firm's body of work, which may or may not resemble yours. The 80% figure is Dram's own estimate; the article ties it to no study.
| What the company says | What the coverage doesn't show | |
|---|---|---|
| Scope | Architecture, structure, MEP and interiors from one sketch or PDF | Which jurisdictions and code editions are covered |
| Quality | Every change checked against local codes | Accuracy or error rates on submitted sets |
| Training | Eight years of approved sets from one design-build firm | How close those projects are to yours |
| Proof | Early customers using the second version | Named customers or independent testing |
| Responsibility | A licensed professional reviews and seals the set | How the software supports that review |
You've probably been pitched software that demoed beautifully and then stalled on your third real project, so your skepticism here is pattern recognition. Each of these gaps could close as customers report back. Until they do, you can prepare for the version where it works by deciding what your review has to become.
Why does faster drafting make your review the job?
The standard behind your seal lives mostly in your head. You know which details pass code and still fail the client: the stair that meets every rise-and-run limit and feels wrong underfoot to a client in their seventies, the threshold that floods in the first heavy rain. A drawing set you drafted yourself carries that knowledge because you made each decision on the way through.
When a set arrives generated in minutes, you get it without that history. One change updating every sheet is the selling point, and it also means one wrong assumption can appear on every sheet at once, drawn as cleanly as everything around it. You get the drafting hours back. You keep the responsibility for what's in the set.
Reviewing a set you didn't draw takes more deliberate attention than reviewing your own, whether a junior drafter or software produced it, and the errors you'd normally catch by drawing them get caught only if your checklist names them.
What should you ask before trusting a generated drawing set?
Ask the vendor, or any tool that makes a similar claim:
- Which jurisdictions and code editions does the checker cover, and how does it tell you when a project falls outside them?
- What's the error rate on sets that went to permit, and who measured it?
- Whose drawings trained it, and how close are those projects to the work you do?
- When the code check and your reading of the code disagree, what does the software show you?
- After a single edit, can you see every sheet it changed?
Then ask for a trial on a project you've already permitted, and compare the generated set against the one you sealed. You already know where the hard decisions were on that project, so you'll see quickly whether the software found them.
Related questions
Can AI generate permit-ready architectural drawings?
Blueprints AI says its software produces code-checked sets across architecture, structure, MEP and interiors from a sketch or PDF, with a licensed professional sealing them. TechCrunch's October 2026 report included no accuracy figures, named customers or independent tests.
Does an AI-generated drawing set still need an architect's seal?
Yes, by the company's own account. Blueprints AI's co-founders say licensed professionals review the documents and affix their seal before permit submission, so responsibility for the set stays with whoever seals it.
How much of an architect's time goes to drafting?
Blueprints AI co-founder Al Dram estimates about 80% drafting and 20% design, as reported by TechCrunch in October 2026. It's his estimate, and the article cites no study behind it.
What should I ask an AI drafting vendor before buying?
Which jurisdictions and code editions it covers, its measured error rate on sets that went to permit, whose drawings trained it, and how it shows you every sheet a single change touched.
What this changes for a small practice
We read this launch as a preview of where a practice's value is heading: if drafting compresses, the taste and judgment behind your seal become most of what a client pays for. That's the standing argument of Build for Augmentation When Your Value Is Taste, arriving in drawing sets.
This week, before you evaluate any tool, write down what you check before you seal a set: the code items you always verify by hand, the client-specific calls, the details that have burned you before. Ten lines is enough to start. That list is how you'd onboard a junior drafter, brief a consultant, or test a vendor's trial set, and it's useful whether or not you ever buy software. Put it where your team can find it; How to Build the Knowledge Base Your AI Agents Can Actually Use covers where a list like that belongs. Getting that standard out of your head is the first stage of how we work, too.
Working together
Flow State Found works with a limited number of businesses to make their best work their baseline. Maybe you've watched drafting eat the weeks you meant for design, and every tool that promised to help asked you to trust output you couldn't check. We help practices write down the standard behind their seal, so any drafting help gets checked against it. We take on limited engagements, so it starts with a conversation.
Start a conversation