Interview
The 1% Problem
ChatGPT told Samim Safaei his fourth startup looked good, and he says that cost him a year. His fifth company, siift, is built to show founders what a chat window leaves out.

ChatGPT told Samim Safaei his startup idea looked good. He spent the next year learning it was a dead end.
It was his fourth company, built with a local partner to automate health and safety paperwork in the construction industry. ChatGPT had just come out, and they used it for research from the start. “It told us, yeah, looks good,” he says. They had revenue within six months.
Then the business needed to pivot, and they found they couldn’t do the thinking on their own. “We had relied on ChatGPT so much and it had kind of outsourced a lot of our thinking and our decisions,” he says. Tariffs and a trade war hit construction, they couldn’t keep up with the competition, and they shut it down. Without the chatbot, he says, they probably never would have started.
What he built next is siift. A founder brings an idea, and within ten minutes it is broken into the parts a business has to work out: the problem, the audience, the value proposition, the channels. Each version is scored on the evidence behind it, from the founder’s own data, the internet and other people, so the gaps show. He started building it in May 2025.
Four companies first
He counts a venture as official when it is incorporated, full time, making money and something he is committed to, “not just a side project that you play with a couple hours a week.” By that count, siift is his fifth.
The first was a fire safety company that licensed patents and landed a big US government research grant. For the second, he teamed up again with his cofounder from the first, a serial entrepreneur who had already taken a company public. They built wireless sensors that helped apartment buildings manage energy and occupancy, and reached several thousand apartments across New England before COVID stalled growth. Safaei left, and the team kept it going.
The third was the one he fell in love with, “which they tell you not to do.” He had been thinking about it since he was a teenager: a device he calls the world’s smartest disco ball, which turned the emotions you hear in music into intuitive visuals, essentially productizing synaesthesia. It drew big waitlists and angel money, including from a former Shazam executive. After almost three years of work and close to launch, pandemic supply chains put some parts a year away. “Music is a very tough business to make money in, and so is hardware, so…” he says with a smile.
After that his fourth was the construction company. By then he had worked in fire safety, smart buildings, music, construction, software and hardware, and was an expert in none of them. What he had was a wide view of how new companies get built, and that became the focus of siift.
No execution filter
To Safaei, the trouble is how easy starting has become. “It’s so much easier now to start a project,” he says. “There’s no execution filter.” You have an idea and you start building, and the tools are designed to engage you and make you feel good. “They’re not really sitting there asking you, is this the right thing for you to do? Should you invest your time in this?” He names ChatGPT, Claude and Gemini, and says he isn’t picking on any one of them.
A chat window makes it harder, he says, because of how little it shows. You get text, maybe a couple of links, maybe an artifact on the screen. You don’t see the reasoning, the logic or where the context came from.
It’s like a black box, like you are seeing 1% of the information.
“That’s not enough for you to make a decision on,” he says. He reads everything these tools give him critically and double checks it. Sometimes a chatbot argues with him while it is clearly wrong, and it takes several passes before it admits the mistake. Some of that is unavoidable, since no model will be right every time. His problem is the format, which makes it hard to tell a real signal from a false positive.
He can push back because he was building companies before ChatGPT existed and knows how a lot of things are supposed to work. He calls that a privileged position, and he worries about younger people who don’t have it. “The chatbot kind of can trick them a lot easier.”
Signal and noise
Telling the two apart, he says, takes both intuition and technical knowledge, “but the intuition is still in some ways kind of technical.” He learned this while building novel fire radiation sensors at his first startup. “If you look at a raw signal, it’s like it’s chaos,” he says. “You have no idea what’s true or what’s not.” What gets you through is tactics, best practices and known scenarios to watch for, which give you rules that reliably filter out the noise. The filtering is technique. Knowing what to filter for, and when, he says, is “a lot more intuitive.”
Founders get plenty of signals that feel like proof: friends who say it’s a great idea, daily signups, LinkedIn likes, a research paper on the space, a first payment. Safaei doesn’t rank them as a list. He sees validation as “a progressive and iterative process,” where you narrow in and raise the bar each round.
The first bar is whether the idea makes basic sense. Can you say why it should exist? The reason doesn’t have to be a good one yet, and it will almost always change. “If the only reason is, oh, I think it’s cool, or I really like this, that’s not enough.” Next is urgency. Everyone has a million problems, he says, and “people usually only pay money for their urgent problems.”
After that comes fit and money. Do customers want something like your solution, will they pay for it, and can you make a profit, even a small one? If a customer won’t pay anywhere near enough, or keeping them happy takes too much manual work on your side, he says the customer, the model or even the problem might be wrong. The last rung is scale: repeat purchases, referrals and what it costs to keep a customer.
“Congrats, you sold a lemonade, you made $2,” he says. “But are you gonna be sitting at this lemonade stand selling every day for the rest of your life to make, I don’t know, $50 a day? Probably not.”
Landing pages, waitlists and social media engagement have their place, but often, he says, they are vanity metrics. They can show that people see the point. They don’t show urgency, willingness to pay, or whether you’re reaching the right people on the right channel. “As usual, cash is king,” he says, “but you have to kind of just take it one step at a time.”
He suggests tracking one to three key metrics at most: one that shows customers are getting value, an engagement or satisfaction measure, and later a scalability measure. The set changes as the company grows. At the very start, the metric can be as simple as signups, site visits, or “how many people even respond to my message when I ask them to talk about this thing.”
That is the process siift tries to measure. Because every version of a problem or an audience is scored on evidence, a founder can see when a different problem has more support than the one they believe in, or when they’ve thought hard about the value proposition and not at all about the audience. He has made a related mistake a lot himself: a big tech plan full of features and no clear understanding of the problem, “a solution in search of the problem.”
Asked why a founder shouldn’t just ask Claude to rate an idea, he points to three things. First, siift connects to a founder’s real metrics, like revenue, and works toward moving them, while a general chatbot, in his view, is “trying to make you spend credits.” It also applies the same frameworks every time, like the Lean Canvas and the build, measure, learn loop. And it keeps a visible record, so teammates and advisors can see why a call was made. “If you fast forward a week or a month, are you going to scroll up 4,000 messages and try to find why it did something?”
Fountain of youth
Safaei is wary of venture capital as the default path for most entrepreneurs. He has VC friends and calls them “fine people in general,” but says the VC business model runs on moonshots, with funds expecting 90% of a portfolio to not return enough. That makes them highly speculative, and often misaligned with a founder’s or business owner’s thinking. Angels and family offices behave differently, he says: “They think more long-term and put more weight in relationships.” He prefers angels, and sees himself becoming one.
After over a decade of building startups, Safaei keeps starting companies anyway because “it is good for my mental health.” Startups are stressful and can consume your life, but they force a beginner’s mind and continuous growth. In a 9 to 5 job, he says, “you feel yourself getting slower, you feel yourself stagnating.” He jokes that entrepreneurship is the fountain of youth.
He doesn’t expect siift to be his last company. He calls himself a visual learner, and a big part of siift is getting information out of the text chat and into something you can see.
Someday, he says, he would like to go back to the disco ball.
“People usually only pay money for their urgent problems.”