Scott Reynolds left the drafting table, called his engineer brother, and built UpCodes - the search engine and AI Copilot that 800,000 building professionals now lean on every month. Then he went to court to keep the law free.
Most architects have the same complaint. Somewhere between the first sketch and the permit, the building code shows up - thousands of clauses, cross-references, exceptions to exceptions, spread across dozens of jurisdictions that each edited the same document a little differently. You lose hours to it. You grumble. You move on to the next project.
Scott Reynolds did not move on. He was working as an architect at Kohn Pedersen Fox in New York, a firm known for skylines, when the code kept getting in the way of the work. He had trained in the discipline - a Bachelor of Architecture from Syracuse - and he had already put in time on the other side of the planet, at the firm Hassell in Hong Kong. He knew the job. He also knew the part of the job everyone quietly hated.
So he called his brother.
Garrett Reynolds was a software engineer at PlanGrid, one of the first companies to drag construction paperwork onto a tablet. The two of them started meeting to hunt for a startup idea. Garrett acted as the reality check, poking holes in one pitch after another. They cycled through options. The one that survived was the least glamorous thing in the room: building codes.
Reynolds talks about building codes the way an engineer talks about bad software. Codes are not written once and maintained cleanly. They accumulate. A provision gets appended, then an exception, then a local amendment, then another - decade after decade, never refactored. He calls the result a "rubber band ball" of code sections. Technically complete. Practically unreadable.
The consequence is not just annoyance. It is ambiguity. And ambiguity in the law is a design flaw.
That framing is the whole company in a sentence. Reynolds does not treat the code as a fixed fact of nature that professionals must memorize. He treats it as a broken interface between people and the rules that govern what they build - an interface that software can fix.
Over the years Reynolds has described UpCodes three ways, each borrowing a tool people already trust: "Kayak for building codes" for search, "Grammarly for construction" for the AI that catches errors as you work, and "TurboTax for code" for where he wants it to go - answer a few questions, get plain-English guidance.
UpCodes started, around 2016, as a free, searchable database of the codes. That alone was useful enough to pull in tens of thousands of users in its first year. The brothers went through Y Combinator. Then, in June 2018, they shipped the part that made people pay attention: UpCodes AI, a Copilot that scans a building model and flags violations in real time.
The press settled on a shorter label: the spellcheck for buildings. During its private beta, the tool caught roughly 27 code violations per project - the kind of misses that, discovered later by an inspector, mean rework, delay, and cost. Reynolds and his brother both came from a data-heavy background, having worked with large 4D data sets including diffusion MRI, and that instinct - turn a messy pile of information into something queryable - runs straight through the product.
Putting the codes online for free was the point. It was also the problem. In 2017 the International Code Council, the nonprofit that writes the model codes most US jurisdictions adopt, sued UpCodes for copyright infringement. The claim: the brothers had posted the ICC's codes and derivative works without permission.
Underneath the filing sat one uncomfortable question. Once a code is adopted into law by a government, can a private organization still own the text of that law? UpCodes bet the answer was no. It was a small startup arguing, in effect, that access to the rules everyone must follow should not sit behind a paywall.
US District Judge Victor Marrero found that UpCodes' use of the codes qualified as fair use - transformative and educational - after a multi-year battle. The decision supported the principle that adopted building codes belong in the public domain. The ICC followed with further claims, but the core copyright fight went Reynolds's way.
Reynolds is candid about how hard it is to raise money for construction software. Most founders pitch investors who already understand their market. He did the opposite - walking into rooms where the ecosystem itself needed explaining before the product could.
He raised anyway - early angel money, backing associated with Foundation Capital, and later rounds bringing total funding to roughly $7.7 million, with the most recent raise around August 2024. He sits on the Forbes Technology Council and carries the Y Combinator alumni badge. But the through-line is not the fundraising. It is a stubborn read on where value hides: in the boring problem the whole industry had learned to tolerate.
There is something almost contrarian about it. The obvious startup ideas are the ones everyone can see. Reynolds and his brother went the other way, toward the thing so tedious that talented software engineers had ignored it for decades - which was exactly why it was open.
Search was step one. The AI Copilot was step two. What Reynolds describes next is a guided workflow - the TurboTax comparison - where you enter your project details and the system walks you to the applicable code in plain English. The expertise that used to live inside expensive specialist consultants, handed to anyone with a laptop.
It is a quiet kind of ambition. No one grows up dreaming of building codes. But the same qualities that made Reynolds a good architect - patience with detail, a tolerance for rules, an eye for where a system fails the people using it - turned out to be the exact qualities the problem needed. He did not invent a flashy market. He found a broken one, and he read the fine print nobody else would.
The building code is still a rubber-band ball. It is just a searchable one now.