OPT   Operator's Take

A few years ago, building a working software product took real money. You hired engineers with years of experience, paired them with designers, and paid for the months it took them to get something out the door. Depending on the vision, that meant tens of thousands to millions of dollars, which usually meant investors or a bank. The cost forced most teams and investors to get serious about what they were building before they built it.

That barrier is mostly gone. A person can now talk to AI over a weekend and launch a product that works.

Being able to build something in a weekend doesn't mean you should, and we think a lot of teams are confusing the two. Hiten Shah made a related case this month: a founder who can suddenly build almost anything will need product marketing more, because every competitor gets the same speed and buyers already have a comparison in their heads before you explain yourself. We agree, and we'd put it more broadly. AI helps with the rest of the work too: figuring out whether you have product-market fit, who needs the product, whether enough of them will pay, and how to set it apart through brand, positioning and marketing. That work still takes time to do well, and AI shortens it far less than it shortens the product build.

When we say shipping got easier, we mean producing a viable product: a codebase that runs, an app that functions, something you could put in front of a market and sell. Viable is a low bar. It doesn't mean good, and a lot of trouble starts when a team treats done as if it meant done well.

Design is often where that gap shows first. At Lenny's summit in September, Stripe's Katie Dill argued that as products get easier to build, interfaces drift toward the generic, and that good design still depends on a clear point of view, a real understanding of users and attention to detail. None of those come with the code.

We've watched this happen inside an operation we've been close to. A subscription product launched its first version for a narrow, expert audience. To reach more people, the next version added far more choice, and it shipped as quickly as the first. New users found it confusing. Nobody had settled who the product was for, so the design had no basis for deciding what mattered most on the screen, and the result was a messy, convoluted interface that tried to be all things to all people.

Every version was easy to build, and each one rested on an assumption no one had tested with real customers. It's an easy trap to fall into, because the speed is real and it feels like progress.

When building was expensive, the cost acted as a natural point of discipline. Nobody wanted to spend money on the wrong product, so teams thought harder before they started.

AI lowers the barrier to entry and raises the floor for what counts as a viable product. More people can build one now, so more products are competing for the same customers. To stand out and win market share, a company has to raise the bar on product quality, positioning, branding and marketing.

That makes progress easy to misread. Things are being built and shipped, which looks like momentum. The better measures are whether the product is getting better, whether it solves the user's problem more effectively, and whether it adds value that didn't exist before. Answering those takes a clear definition of success and the discipline to measure it, which is much harder than counting what shipped.

People who aren't practiced at defining an audience tend to say their product is for everybody. They know that isn't true, and they still can't name who it's for or who it isn't for.

There's a money trap here too. When you're thinking about revenue, narrowing the market feels like leaving money on the table. In practice, being precise about who your product serves is how it ends up being successful. And if the need only exists for a handful of people who won't pay much to solve it, you don't have a business yet, however fast you build it.

Doing this well means getting specific about which group in the market you serve, what drives them, what they spend on the problem today, and what they'd pay to solve it better. It also means naming the people who look like customers and aren't. That's a business decision, and someone has to make it.

Two groups have to settle it. Owners or investors, who have the capital at stake, need to agree the audience is big enough to make money against what the product will cost. Then the answers go to everyone who will build and run it: product, engineering, design, marketing and sales.

To start: write one paragraph on who your product is for and one on who it isn't for, and put both in front of whoever is paying for the build.

Speed is a real advantage, and AI earns its place at this stage. Share a rough observation about a market and ask where the gaps are. Ask for audience profiles covering who would use the product, what they earn, what they care about and how old they are. Ask it to find competitors and break them down. You'll have a solid first pass in an afternoon.

What you can't do is assume that because the first pass was fast, the rest will be. Some of the output will be right and some won't. A lot of it will sound logical and accurate with no data behind it, and that part should be held as a theory until you can test it. Check whether these are the right competitors, and whether the example you gave the AI was even correct. A long research report with clean charts can still rest on a wrong premise.

Shah ran a version of this test. He asked Claude why Linear loses deals to Jira using only public web research, with no buyer interviews or deal records, and it produced plausible reasons anyway. Once he gave it a win/loss skill, it treated those reasons as hypotheses and listed the evidence the team would need before making a call.

When every competitor has the same speed, the quality of your thinking is what sets you apart. Working through AI research with real critical thinking takes time. Sometimes it takes weeks, and sometimes you learn something that changes the plan. We all know the saying "haste makes waste," and now AI makes haste very easy.

To start: take the last AI research doc your team relied on and mark each claim as backed by your own customer data or as a theory still to test.

The profiles you built are also test material. Turn each one into a testing script that walks a first version through what that person would actually do (sign up, make a purchase, look for the feature they care about), then have AI run the script as that person, one profile at a time.

Without AI, this meant recruiting test groups, paying them and running sessions. Now you can run a version of it from your desk, or your phone, and get results quickly. Persona testing is a great checkpoint before real users; it doesn't stand in for them.

Treat the findings the way you treated the research. Change what clearly needs changing. When a finding is complex and you're not convinced, log it instead of spending two weeks on it, and watch for user behavior that confirms it once the product is in the market.

To start: turn each customer profile you've written into a short test script, and have AI run your prototype once as each.

Good entrepreneurs know an idea is an educated guess, even when data backs it. Plenty of successful businesses found their real opportunity only after customers responded to an early version. The point of all this preparation is to launch a first version with clear intent, then learn from it.

Be clear about what the product is and who it's for, build that version fast, and decide how you'll measure success. Then watch whether people buy, where they get stuck, and whether they wanted the thing you built or something you were only half offering. Prove it works with a limited launch before spending heavily on acquisition, then go wider.

By then, AI holds your audience profiles, your positioning and the history of every prototype. That makes it a far better partner on messaging and creative, because it knows who it's writing for.

To start: before your next build cycle, write down the one signal from real users that would tell you the current version is working.

When building gets this cheap, the pressure moves to the work that decides whether anyone cares about what you built: positioning, brand, messaging, and design that reflects what a specific person needs. AI can support every part of it. Doing that work well still depends on human judgment, creativity and a real standard for quality. That takes time. It always has, and it still does.

Helping teams set up that kind of process, with AI at every stage and people making the calls, is the work we do with the brands we work with.