UpCodes.

Here is a fact about building codes that is either boring or fascinating depending on how much you have to deal with them: the rules that keep a building from collapsing, catching fire, or trapping people in a stairwell are not written once and posted in a public place. They are written by a standards body, adopted - with edits - by thousands of separate states, counties, and cities, amended locally about 160,000 times, and then, in many cases, sold back to the architects who are legally required to follow them.

If that sounds like a business opportunity dressed up as an inconvenience, you are thinking like the Reynolds brothers. Scott Reynolds was an architect at Kohn Pedersen Fox in New York, having worked across jurisdictions as far apart as Hong Kong and Manhattan. His brother Garrett was an engineer - a physics and machine-learning background from UCLA, and a stint building software at PlanGrid. One of them knew exactly how painful code research was. The other knew how to make software eat that pain. In 2016 they started UpCodes.

The premise is almost aggressively simple: take every building code in the United States, put it in one place, cross-link it, keep it current, and let people search it like the internet taught everyone to expect. The complexity is entirely underneath. There is no single "building code." There is the model code, then the adoptions, then the amendments, then the version that applies to your specific project in your specific city in the year your permit was pulled. Getting that right is a data problem disguised as a reference book.

The free version of UpCodes is the front door, and it is genuinely free: type a question - occupancy load, guardrail height, egress width - and get the relevant section, in the right jurisdiction, cross-referenced to everything it touches. For the roughly 650,000 people who visit each month, that is often the whole product, and that is fine. Giving away the search is how you earn the habit.

The paid product is where UpCodes turns a reference tool into a workflow. Teams annotate codes, share compliance notes across a project, bookmark sections, and pull the whole thing up on a phone while standing on a job site. In an industry that still runs on marked-up PDFs and institutional memory, "everyone looking at the same version of the same code" is not a small feature.

Then, in 2023, UpCodes did the thing every software company did in 2023: it added AI. But it did it with a constraint that is worth pausing on. Copilot, built on GPT-4, answers code questions - summarizing sections, explaining them, running calculations, building checklists. The trick is that UpCodes "fences" the model into only the codes that apply to your project's location and permit year, parsing over five million sections and 160,000 amendments to decide what the model is even allowed to see. It cites its sources with links back to the exact section. In a field where a confidently wrong answer about fire ratings is a genuine liability, refusing to let the AI free-associate is the product.

Now for the part that makes UpCodes more than a nicely designed database. In 2017 the International Code Council - the body that writes the model I-Codes that most of the country adopts - sued UpCodes. The argument, roughly: you cannot republish our codes without our permission, even the ones that have been written into law.

This is a genuinely hard question, and it is the kind of question Matt Levine would enjoy: when a private organization writes a set of rules, and a government adopts those rules as binding law, who owns the words? In May 2020, a federal judge in the Southern District of New York offered an answer favorable to UpCodes, finding that posting adopted codes served a transformative purpose - "disseminating enacted laws for public awareness" - and was fair use. The dispute has continued through appeals; as recently as 2026 a Third Circuit ruling held that UpCodes' publication of incorporated standards is likely fair use.

The strategic point is that UpCodes' moat is not only its software. It is partly a legal principle - that the law you are required to obey should be something you can read - and the company has spent years defending it in court. That is an unusual asset. Most startups worry about competitors. UpCodes has been busy establishing that the raw material of its business belongs, in a meaningful sense, to the public.

UpCodes went through Y Combinator and picked up backing from Foundation Capital early. The rounds since have been modest by 2020s standards, which fits a company selling into architecture and engineering - industries that move deliberately and reward tools that earn trust before they ask for a credit card. The through-line in its cap table is construction-tech credibility: the founders of PlanGrid, where Garrett Reynolds once worked, are among the investors.

The customer is anyone whose job depends on getting the code right: architects, structural and MEP engineers, contractors, building product manufacturers, and the code officials on the other side of the plan check. The competition, honestly, is not glamorous - it is the International Code Council's own digital codes, a stack of PDFs, a $400 printed book, and the senior person in the office who "just knows" the requirement. UpCodes is betting that a searchable, cited, always-current answer beats all of those, and about 650,000 monthly visitors suggest the bet is landing.

What makes the company worth watching is the shape of the problem it chose. Building codes are unglamorous, fragmented, slow to change, and legally contested. That is exactly why they were still trapped in binders in 2016 - and exactly why turning them into software, and then into something you can ask a question, is worth doing. UpCodes did not invent a new category of desire. It took a chore that everyone in construction already hated and made it take seconds instead of an afternoon. In an industry that runs on liability, that is not a small thing to sell.