For the past few weeks I've been calling general contractors, architects, and subs. Asking questions. Trying to understand the actual mechanics of how information moves through a construction project.
One thing kept coming up that I can't stop thinking about.
An architect designs a building. They model it in Revit. They have every wall assembly, every footing dimension, every door schedule. The quantities exist. The specs exist. The spatial relationships between systems exist. It's all there, sitting in a 3D model that took months to build.
Then the architect exports a set of 2D drawings. Flattened PDFs. The model becomes paper.
Then a GC receives those drawings and starts the bidding process. And someone, whether it's an in-house estimator or a third-party firm, has to go through those flattened documents and manually extract the quantities that already existed in the model upstream.
One GC I spoke with told me they hire a third-party firm to handle this. "We'll hire out some firm out of India. And then they get back to me basically within 24 hours, and then we evaluate our own quantities as well." The cost is roughly $2,500 per project. Two separate teams recreating information that the architect already had.
When I asked people why this happens, the most common answer was some version of "that's just the way it is."
That answer is true. But it's also incomplete.
Why the Model Stays Locked Up
I had two conversations recently, one with an architect and one with a builder who's worked in and around different parts of the industry for decades, that helped me understand the deeper reason.
The architect explained something I hadn't fully appreciated. The plans and the Revit model are not the same thing.
The plans are the contractual source of truth. Every note, every detail, every schedule on those sheets has been reviewed and stamped. The Revit model is a representation. It's built alongside the plans, but it doesn't always match perfectly. A drafter might miss updating a footing. The structural section might not reflect a late revision.
He walked me through the risk. "What if we have something screwed up in there and then they come back and said, well, you forgot the roofing in three of your four rooms?" His point was that the notes and details on the plans are always correct, but the 3D model might have discrepancies that the architect never intended anyone to bid from. "The plans are always right, but the model may not always be perfect."
So architects don't share their Revit models. Not because they're being difficult. Because sharing them creates risk they're not willing to absorb. As he put it, "That's why the plans are there. Just bid off the plans."
And when I asked about the process of building those plans, whether there's a standard workflow for how the data gets entered, his answer caught me off guard: "Just kind of depends on who was drafting it that day."
That's not a knock on his firm. That's the industry. Some firms build the Revit model first and export drawings from it. Some build drawings directly and the model is secondary. Some have dedicated draftspeople who fill in the specifications that the design architect doesn't detail. The path from design intent to final drawing set is different everywhere.
The Intentional Gaps
The second conversation added another layer. This person has been in and around construction his whole life and has seen how different parts of the industry interact up close.
He pointed out that architects don't just withhold the model. They intentionally leave information out of the drawings themselves. Pour stops. Shear studs. Steel deck specifications. Metal stud framing details. Insulation types. Wall assembly components. Door hardware. Things that exist in real buildings but don't show up on plan sheets.
The logic, as he explained it: less information conveyed means less liability assumed. More detail on the drawings means more surface area for someone to point at later and say "you told us X, but it should have been Y." So architects keep the scope lean and push the responsibility for filling in the blanks downstream, to the subs.
An experienced mechanical contractor knows what insulation type goes on a chilled water line even if the drawing doesn't specify it. A structural steel fabricator knows the connection details that aren't drawn. That trade knowledge is expected to fill the gaps the drawings leave open.
This works, mostly. Until a sub bids on what's shown and the GC expected them to include what wasn't shown. That space between what the drawings say and what the project actually requires is where most disputes live.
The Scope Division Problem
Even if an architect could hand over a clean set of quantities, the GC would still need to break those numbers apart by trade.
An architect might know the total concrete quantity for a building. But concrete on a construction project doesn't belong to one subcontractor. The structural concrete is one trade. The mechanical pads are another. The site work concrete is another. Flatwork, curbs, and sidewalks might be a fourth.
The architect thinks about the building as a design. The GC thinks about the building as a collection of scoped work packages that need to be priced by different companies. Those two mental models don't map cleanly onto each other.
So even in a world where the data flowed perfectly from architect to GC, someone would still need to do the translation work of breaking a unified design into trade-specific scopes. That work requires construction knowledge, local market knowledge, and an understanding of how subcontractors in your area actually divide responsibilities.
There Is No Unified Process
I started these calls expecting to find a technology gap. Something that better software could bridge. And there is a technology gap. But it sits on top of something more fundamental.
The construction industry doesn't have a unified process for moving information from design to bid. And the fragmentation isn't an accident.
Architects limit what they share because sharing more means owning more risk. GCs accept the manual extraction work because they don't trust anyone else's numbers anyway (that same GC who pays $2,500 for outsourced quantities still cross-checks every number internally, "bid by bid what the sub said and what your company determined"). Subs fill in the blanks from experience because that's what their trade knowledge is for.
Everyone has adapted to the inefficiency because the alternatives require someone to accept liability they don't currently carry.
That's the real reason a GC pays thousands of dollars to have someone re-extract quantities from a PDF that was exported from a model that already had the quantities. The technology to share that data exists. The contractual and cultural framework to share it safely does not.
A lot of construction tech is built on the assumption that the industry's inefficiencies are caused by bad tools. Some of them are. But the biggest ones are caused by misaligned incentives between parties who are all trying to protect themselves. And until you account for that, better tools just move the friction around without removing it.
"That's Just the Way It Is" Usually Means Innovation Is Coming
Every industry that's ever said "that's just how it works" eventually got disrupted by someone who figured out a better way. And I think AI is going to crack huge chunks of this problem open.
Think about what's actually possible right now. A GC determines the scope from the drawings. Instead of paying $2,500 and waiting on a team overseas, multiple AI models extract quantities from different angles and cross-check each other for discrepancies. The contractor reviews a handful of flagged items instead of re-verifying every number from scratch. Scopes get handed off to subs with quantities already attached, so the sub spends less time on their own estimating and there's less back and forth between GC and sub over what's included. Cleaner scopes mean cleaner bids. Fewer surprises. Less rework during construction.
None of this eliminates the judgment calls. Someone still needs to decide how to split concrete between trades. Someone still needs to catch the things the drawings don't show. But the hours and hours of manual re-extraction, the $2,500 per project, the two teams independently recreating the same information? That part doesn't need to stay manual forever.
The liability question doesn't go away either. Architects still won't share their Revit models, and there are good reasons for that. But if AI can extract what it needs from the plans themselves, the flattened PDFs that everyone already has access to, the model-sharing problem becomes less relevant. You're working with the contractual source of truth, the same document the architect is comfortable standing behind.
I believe the pieces are falling into place for this. The question now is how. How do you build it in a way that respects the existing workflows instead of fighting them? How do you make it trustworthy enough that a contractor will actually rely on the output? How do you get adoption in an industry where the default answer to everything is "that's just the way it is"?
That's what I'm working on at Margo And that last question might be the hardest one.
I'm writing about what I'm learning as I build this company. If this resonates, I'd love to hear how your firm handles the gap between what architects provide and what your team actually needs to bid a project.