Here's an uncomfortable thing for everyone currently selling "AI." Your competitor can call the same model you do. Same weights, same API, same price. The intelligence is rented, and the landlord rents to anyone with a credit card.
So the output converges. Same glossy look. Same confident caption. Same faint AI smell — that flat, plausible, made-by-no-one quality you can now spot at a glance, the way you can spot a stock photo before you've consciously registered why. When everyone paints with the same brush, everyone paints the same painting, and the market gets very good, very fast, at scrolling past it.
Which means the model was never the moat. It's infrastructure — necessary, cheap, and shared. The moat is taste: knowing what to make, what to kill, and what actually sounds like you instead of the median of everyone else's prompt. Taste is a human standard, a point of view the system has to answer to before anything ships. It's the one part that doesn't come bundled in the API, because it can't. It has to be decided, on purpose, by someone willing to say no.
Here is the mechanism, not the metaphor. A system with volume and no taste just produces more of the wrong thing, faster. So the real work isn't generating anymore — generating is free now. The real work is filtering. Somebody has to look at everything the agents made this week and ask the boring, specific questions: does this sound like us, or like anyone? Would it survive without the logo on it? Is this teaching something, or just performing the shape of teaching? Most outputs fail one of those questions. That isn't a flaw in the process. That is the process. Nine hundred and ninety get killed so ten can be kept, and the ten that survive are the ones with a fingerprint on them.
This is why we build the gate into the system instead of bolting it on after the fact. The agents carry the volume — that leverage is real and it's the whole point of hiring them. But a human holds the line on what's worth making, every single time, before anything goes out the door. Skip that step and you haven't automated your marketing. You've automated your averageness, and you're paying compute to do it faster than your competitors do theirs.
So if your AI strategy is "we use the same model, but more of it," that isn't a strategy. It's a faster way to look like everyone else, at scale. Point the same intelligence at a real point of view — one you're actually willing to enforce, even when enforcing it is slower — and it compounds into something a competitor can't just subscribe to. Point it at nothing in particular, and it just makes more nothing, at a speed that used to feel like progress.
Takeaway: models are rented by everyone. Taste is owned by you. Build the system around the judgment, not the tool. ✱
