Sometime around 2010, Reid Hoffman handed a generation of founders permission to ship junk:
“If you're not embarrassed by the first version of your product, you've launched too late.” — Reid Hoffman, reflecting on the line he coined
It became one of the most repeated sentences in startups, and for good reason. It anchored the whole Lean Startup canon: build a minimum viable product, put it in front of real people, measure, learn, repeat. Don't polish a thing nobody wants. Ship the smallest version that tests your hypothesis, and let reality tell you what to build next.
It was excellent advice. For its time. The trouble is that the time is over, and most of us haven't updated the playbook.
The assumption nobody examined
The MVP was never really about minimalism as a virtue. It was a hedge against cost. Building software used to be slow and expensive — months of engineering, a real team, a real budget. You didn't want to sink half a year and six figures into something the market would shrug at. So you shipped the smallest thing that could answer one question — do people want this? — and you kept the rest in your head until the answer came back yes.
Notice what's smuggled into every argument for the MVP: building the full thing is expensive, so don't — not until you're sure. “Minimum” and “viable” were both compromises. Not ideals. Concessions to the price of building.
Reprice building, and the whole logic comes apart.
AI just repriced building
In early 2025, Andrej Karpathy — who helped build the modern AI stack — described his own workflow and accidentally named an era:
“Fully give in to the vibes, embrace exponentials, and forget that the code even exists.” — Andrej Karpathy, coining “vibe coding”
Whatever you think of the phrase, the underlying shift is real and it is enormous. Work that used to fill a sprint now fills an afternoon. Adding a feature is no longer a roadmap negotiation; it's a prompt. Polishing a rough interface isn't a designer-week; it's an evening. As one Inc. piece put it bluntly, AI means you don't need a minimum viable product — you can build the real thing in the time it used to take to fake it.
When the cost of building the fuller, better version collapses, the reason to ship the bare version collapses with it. That's the part that's genuinely dead. Not iteration, not customer feedback — the specific move of deliberately shipping something worse than you could, to save money you no longer have to spend.
But “fast” is not “good.” This is where the hot takes go wrong.
Plenty of “the MVP is dead” posts stop here and declare victory, which is a mistake. Cheap-to-build is not the same as done. AI hands you working code, not necessarily good code — one 2025 analysis of AI-assisted development found generated code carried measurably more defects than human-written code, and the gap between “it runs” and “it holds up” is exactly where most AI-accelerated products quietly fall apart.
So the bottleneck didn't vanish. It moved. It used to sit at the keyboard — can we build this? Now it sits on either side of the keyboard: is this worth building, and is it actually any good? Speed is solved. Taste is not. Judgment is not. Knowing which features earn their place, and having the standards to tell when something is finished, is now the entire job rather than the reward at the end of a long build.
You can't vibe your way past that. But you also no longer get to hide behind “we'll polish it later.”
The part that actually matters: expectations just moved
Here's the consequence almost nobody prices in. If everyone can now ship fast and polished, then polished stops being a differentiator and becomes the cost of entry. The engineering teams watching this closely already see it coming — 2025 was the year of AI speed; 2026 is the year of AI quality.
Users don't grade on a curve. Nobody opens your app thinking “not bad, for a small team.” They compare it — instantly, unconsciously — to the best thing they touched yesterday. And when the best things get more polished, “good enough” gets redefined upward automatically, without anyone deciding it. The floor rises under your feet.
Which is why the MVP is being quietly replaced by the Minimum Lovable Product — and the reason isn't a cuter acronym. It's that “viable” no longer clears the bar. A merely viable product in 2026 is an invisible one.
So what do you actually do?
Four things, in order:
- Validate the problem, not the product. The one thing AI did not make cheaper is being right about what people want. The Lean Startup's real engine — build, measure, learn — is fine. It's the word “minimum” that died. Keep talking to humans; just stop using cost as your excuse to ship them something thin.
- Ship the lovable version, because now you can. The hours you'd have spent defending a bare-bones MVP in a roadmap meeting? Spend them on the product. The math finally favors doing it right the first time.
- Pour the surplus into the 20% AI can't do. The last mile — the taste calls, the edge cases, the way it feels — is where products win now, precisely because it's the only part that hasn't been commoditized. Generic is free. Considered is not.
- Raise your own bar before your users do it for you. Assume a competitor shipped the polished version of your idea this week. Because they might have — in an afternoon.
We build this way on purpose
This isn't a thought experiment for us. Berkeley Labs is a small studio that ships real apps — Kiron, which takes an AI-generated workout plan and turns it into a trainer that actually tracks it for you, and B-Side, which hand-picks five new albums a week with a reason you'll like each one. Neither started life as an embarrassing first version. We skipped that step, on purpose, because the tools let us — and spent the saved time on the details that make people stay. (And if you're wondering whether it's even worth building products on top of the big AI models at all, that's a fair question we take on separately: no, ChatGPT won't eat every app.)
That's the whole shift, really. Moving fast used to be a competitive advantage. Now it's table stakes — everyone can move fast. The advantage went to the people who use the speed to be good. The embarrassing first version isn't brave anymore. It's just slow taste with a head start it no longer has.
Anyway — for a studio run by dogs, we're weirdly particular about what we drop at your feet. Nobody wants the slobbery, half-chewed stick. We fetch the good one, or we don't come back at all. 🐾
