Anyone can ship now. Taste decides who wins.

Two years ago, getting a working product in front of customers took a team and a quarter. Today one person with a coding agent can do it over a weekend. Andrej Karpathy gave the new way of working a name in February 2025: vibe coding, where you describe what you want, accept what the model writes, and "forget that the code even exists".

That is a real change, and we use it every day. But look at what it changed. It made the making cheap. It did nothing to the choosing. And the choosing was always where products lived or died.

Time-to-market is now everyone's

For most of software's history, speed was an advantage you could buy: more engineers, better tooling, a team that had done it before. That moat is drying up. When a quarter of a Y Combinator batch reports codebases that are almost entirely written by AI, shipping fast is no longer what sets a company apart. It is the entry ticket.

So if everyone can ship, what decides who wins?

Taste: knowing what, when and why

In a 1995 interview for the PBS series Triumph of the Nerds, Steve Jobs said the only problem with Microsoft was that "they just have no taste". He did not mean fonts. He meant the ability to look at a hundred possible things and see the one worth doing, and the ninety-nine worth leaving out.

Taste, the way we use the word here, is three judgments:

• What: which problem, for whom, and what the product should refuse to do.

• When: why now, and not two years ago or two years from now.

• Why: the reason a person would change their habits for this, in their words, not ours.

An AI model can write all the code for a wrong answer to any of these just as fast as for a right one. It will not tell you which one you asked for.

Hero or zero

Give two teams the same model, the same budget and the same weekend. One ships a product people keep using. The other ships a product nobody needed. The tools were identical. The difference was taste and intuition: what they chose to build, and what they chose to leave out.

That difference used to be hidden behind execution. A team with average taste and great engineers could still win on speed. Now that execution is cheap, taste is exposed. It is the variable that is left.

And no, this doesn't mean SaaS is dead

You will have heard the other version of this story: if anyone can build software with an agent, nobody will pay for it, and SaaS is over. We think that gets it half right. What AI ends is software whose only value was being hard to build. Software that carries real judgment, about what to include, what to leave out and how to behave on the worst day, is worth more than ever. SaaS isn't dead. Software without taste is.

What this series is about

Over the next pieces we look at what AI-assisted building really solves, what it doesn't, and how to get better at the part it can't do:

• What vibe coding actually solves, and the problem it leaves untouched.

• Taste as subtraction: the hundred good ideas you say no to.

• Working backwards from the customer, with a 1997 talk that still holds.

• Real artists ship: speed as a tool for learning, not a strategy.

• Intuition is trained, not born.

• Timing: why when matters as much as what.

• A short checklist to run before you write your first prompt.

• The "death of SaaS", and why it is still about taste.

Build fast. It has never been easier. Then spend the time you saved on the question that decides everything: is this worth building?