Unbundling the Architect

• August 4th, 2026

Everyone is asking whether AI will replace architects. They’re asking the wrong question.

By Sandeep Ahuja and Shaun Abrahamson

Here’s a pattern worth noticing. A boring, profitable, vertically integrated industry hums along for decades. Then a technology shows up that’s good enough to do one slice of the work on its own — and the whole thing comes apart at seams nobody knew were there. Banks bundled a dozen products until fintech pulled them apart. Detroit built the whole car until the supply chain ate the value and left assembly as the thin part. The thing that looked like one business was several businesses wearing a trench coat, and the technology just unzipped the coat.

Now it’s happening to architecture — but we’re thinking about it the wrong way. The conversation right now is about productivity: how do we make architects faster, how do we get a 10x architect. The same narrative we saw for software developers. That framing will disappoint everyone betting on it, because a firm was never one thing that gets faster or slower. It’s two completely different businesses sharing a logo and a coffee machine — judgment and production.

Judgment is knowing which of a hundred options is right, reading a site, telling a client the thing they don’t want to hear, taking on the liability of a stamp. Small, expensive, irreducibly human. Production is turning judgment into the thousands of coordinated work products a building requires — zoning analysis, code checks, unit plans, egress diagrams, schedules of materials. Large, labor-intensive, and for the whole history of the profession, the thing that actually employed most of the people in it.

These two have always been bundled, because they had to be. You couldn’t get the production without hiring the bodies, and you couldn’t bill for the judgment without wrapping it in production. The hourly model exists precisely because the two were inseparable: you sold time because time was the thing you had to buy. And the buyer was almost never the person who’d end up in the building. It was a developer, who wasn’t really buying hours — they were buying a permitted, financeable project. Hours were just the toll.

AI doesn’t make this bundle faster. It breaks it.

How is it that construction productivity has gone sideways while every other industry compounded? CAD and then BIM delivered real production speedups — so why didn’t they show up in the numbers? Because taking less time in a business that bills for time doesn’t make sense. It’s a threat to revenue.

What’s new isn’t the capability. It’s that the capability finally got good enough to detach production from judgment entirely. There’s no need to buy production anymore — it’s fast and cheap, and iterations are nearly free. But judgment? That matters more, and it’s worth paying for, because judgment is where the returns actually come from: the project that pencils instead of dying, the site that turns out to be worth more than it looked, the building that still works in thirty years.

This isn’t a thought experiment. We’re doing it.

At cove, the production layer is largely automated now: a site address resolves to a parcel, the parcel pulls its own zoning, the zoning constrains the massing, the massing feeds the pro forma, and a code change deep in the set cascades back through everything it touches without a person remembering to chase it. We’ve taken a multifamily project in Atlanta through coordinated design in fifteen days, designed a Tier III data center in thirty, and run zoning, fire-code, and FEMA compliance as continuous checks rather than end-of-project audits.

That’s a 10x productivity improvement — on production. Which is exactly the point. The 10x landed in the half that was never the valuable half.

Treat those numbers as company-reported, because they are. The point isn’t the figures — it’s the shape of the change. The cost of one more iteration has fallen by roughly an order of magnitude, and when iteration gets that cheap, the architect stops producing the answer and starts choosing it. That’s not a 10x architect. It’s a different job, on top of a different company.

Why it matters

Pricing and labor. Two separate businesses can’t be sold as one line on an hourly invoice. Judgment is scarce and should be priced on outcomes; production is abundant and should be priced like software. That’s wrenching for a profession whose economics and training assume the bundle — the junior years that taught judgment through production don’t exist in the same form, and nobody has solved what replaces them.

How fast things get built. The binding constraint on housing isn’t construction — it’s everything upstream: entitlement, design, code, the grind that turns a viable site into a permitted one. That grind is mostly production labor, exactly what unbundling makes cheap and fast. If the slow part gets fast, more of the marginal project pencils.

What gets locked in. Every building is a fifty-year bet, and the decisions made in the first weeks of design commit the built environment to consequences that outlast everyone in the room. When iteration is nearly free, you can test the better version against the default and actually choose — and whoever controls that layer quietly controls a great deal of what gets built and left there.

Which is where this stops being a story about architecture firms

An architect sits at the top of a chain: architect, then developer, then the lender and the land they’re bidding against, then eventually whoever lives or works in the thing. Unbundling changes the cost of the first link. What happens to that saving as it moves down the chain is a completely separate question, and nothing about the technology answers it.

Cheap production should democratize good building. The projects the old labor model priced out — infill, adaptive reuse, the missing middle — suddenly pencil, and a class of building that simply didn’t get made starts getting made. We believe that version is possible. We don’t think it’s automatic.

Because cost savings don’t flow to renters just because someone built a faster design engine. They flow to whoever has pricing power, and in most markets that’s the developer and the land. A decade of climate and proptech companies learned this the hard way: you cut a real cost out of a project, and it gets absorbed one layer up before it ever reaches the ground. Cheaper production democratizes building no more than cheaper steel did. It changes who captures the margin — and left alone, the margin goes to whoever already had the leverage.

So the firms still asking how to make their architects 10x more productive aren’t just answering the wrong question — they’re missing the one that matters. It was never how fast the old thing can go. It’s what the two new things are worth, who they serve, and who decides. That part isn’t settled. It’s being decided right now, by the people building these firms and the people funding them, and most of them haven’t noticed they’re making the call.