What Seven Years Building a Startup Actually Teaches You

There's this moment in Little Miss Sunshine where the delusional dad character keeps making "one more call" for his doomed self-help business, and everyone watching knows it's never going to happen. Greg Law, co-founder and CEO of Undo, watched that scene with his wife after seven years of grinding on his startup. Neither of them said anything. They didn't have to.

Law gave a talk at CppCon 2025 that strips away every comforting myth programmers tell themselves about starting companies. It's the kind of talk that should be required viewing before anyone types git init on their brilliant shower idea. Not because it'll stop you—if you're the type who needs to be stopped, you weren't going to make it anyway—but because maybe you'll see the traps coming.

The Lie You Tell Yourself First

"I had this idea and I wanted to bring it to the market," Law explains about his vision for time-traveling debugging in C++. "The best way to do that was to create a company so I could pursue this vision and I could make programming better. Yeah. Bullshit, right?"

The real reason? Ego. "I think all entrepreneurs have something kind of a bit wrong with them," he says. "We weren't hugged enough as kids or something. I don't know what it is. And we just need this kind of validation."

This matters because the story you tell yourself determines everything else. Law thought he was building a company to ship great technology. Actually, he was building a company to prove something to himself. Once you know that, you can at least work with it.

It wasn't about money—Law is clear on this point. If you're a good programmer in tech, your expected financial return from climbing the corporate ladder beats the median outcome of starting a business, which is "either zero or less than zero." Unless you specifically need to be a billionaire (in which case, sure, go ahead), money is a terrible reason to start a company.

The Part Where Everyone Gets It Wrong

Law had the technical insight in the shower (side note: he also got his PhD idea on the toilet, but phones killed that particular venue for breakthroughs). The core concept for time-traveling debugging: if you want to go backward in code execution, start from a snapshot and replay forward. Simple enough that it fit in his head, complex enough that it might actually work.

He pitched it to Chris, a super-smart former colleague. Chris didn't get it. Why would you need this when stack traces exist?

Law's response at the time: "Chris must be having a bad day."

Law's assessment now: massive red flag. If you can't explain why your thing matters to someone smart in your field, you have a problem that won't get better when you're trying to explain it to journalists receiving a thousand press releases a day.

But here's where it gets interesting. Law found the right co-founder in Jules, someone who immediately got it and replied "I'd love to start a business with you." That partnership lasted over a decade and was "absolutely instrumental" even though it eventually ended. The lesson isn't that co-founders are forever—it's that you need someone you genuinely trust, because disagreements are coming and without that foundation, you're done.

What Programmers Get Catastrophically Wrong About Sales

Before starting Undo, Law was the engineer sales teams brought to meetings. He thought he was doing all the work: writing most of the slide deck, doing the demos, answering the technical questions. The salespeople just did the first two slides and collected the money.

"I had no idea how much goes into getting in that meeting in the first place and then how much on the back of it to actually turn that into, you know, money," Law admits.

He used to think engineering was the hard, sciency thing and sales was fluffy and easy. Then he realized: "With programming, you just have to get the computer to do what you want. With sales, you have to get people to do what you want. And with marketing, you have to get people to do what you want, and you don't even get to meet them in order to do that."

This reframing is kind of devastating if you think about it. Computers are deterministic. People aren't. Making people do things is objectively harder than making computers do things. We just don't experience it that way because we've spent years getting good at talking to compilers.

The Seven-Year Mistake

When Law's former bosses tried to talk him out of starting Undo, they couldn't poke holes in the technical approach. So they pivoted: "Can we just have a chat about this business thing because I think it might be harder than you think."

Law: "No, no, no. Guys, honestly, this is different."

Them: "Yeah, you know, everyone says that."

Law: "I know everyone says that, but this really is different."

It wasn't different.

The launch went exactly how these things go when you're wrong about product-market fit but don't know it yet. Law Googled "how to write a press release," sent one out, warned his hosting company about the impending traffic surge. Nothing happened. Nobody cared. Technology journalists get a thousand press releases a day about amazing new tools.

What followed was seven years of part-time grinding before Undo had enough revenue for Law to quit his day job. Seven years of being that guy from Little Miss Sunshine, except occasionally it actually does work out.

The mistake, Law says, was "misdiagnosis." He thought they were close to product-market fit when they launched. They weren't. "Any strategy is all about the diagnosis," he explains. "If you get the diagnosis of where you are wrong, you're going to have the wrong strategy. You get it right, the strategy just kind of writes itself."

This is maybe the most useful frame in the whole talk. Not "work harder" or "be more resilient" or any of the other platitudes. Just: figure out where you actually are, not where you want to be. The strategy follows from that.

What Actually Helps (And What Doesn't)

Law's advice on seeking help is surprisingly practical. Incubators like Y Combinator are legitimately good (they didn't exist when he started). Government grants and programs exist, though the "desire to help and the actual implementation of help is different." You can cold-email successful people and many will actually reply—some will even buy you lunch.

But beware the "hangers-on"—people who retired early from big companies and now want to advise startups. "Have they done this before?" is Law's filter. Either they've built something themselves or they've successfully coached people who have. Otherwise, "probably no help at all. In fact, probably negative help."

The talk is aimed at programmers who want to build companies or programmers thrust into leadership roles, but it's really about the psychological gap between solving technical problems and solving human problems. One has a compile button. The other requires figuring out what you actually want, why you actually want it, and whether the story you're telling yourself matches the story everyone else is experiencing.

Law spent seven years as the protagonist of a story that maybe wasn't happening yet. Eventually, it did happen—Undo has real customers and real traction. But the question that hangs over the whole talk is the one he asks himself: "Why seven years? Like if I'd known it was going to—"

The transcript cuts off there, but you can fill in the blank. Would he have done it differently? Would he have done it at all? And if the answer is yes either way, what does that say about the kind of person who starts companies?

---

Yuki Okonkwo is Buzzrag's AI & Machine Learning Correspondent