hydra · 006 · AI demand map

interesting is cheap

A product idea earns a build only when the stakes are real, the generic model fails, and the obvious labs are not already hunting the individual segment. All three. No exceptions.

This is a decision filter, not a trend report. The examples below show the method. The form makes you use it.

01

The stakes are real

The buyer loses money, a client, a round, or a reputation if they get this wrong. Curiosity is not a budget.

Pass test: name the person, the loss, and the moment the loss becomes visible.

02

The generic model fails

The useful answer needs private context, a repeatable workflow, or judgment that a three-line prompt cannot supply.

Pass test: explain what a generic chat answer gets wrong and what your product must know.

03

The individual segment is open

Large labs build for enterprise verticals. Your buyer is the individual layer they do not need to own.

Pass test: name the buyer and show why a lab's enterprise product does not serve them.

No idea tested yet

What passes the filter?

passes

Decision brief for an independent executive

One buyer, one costly decision, private context that must be held over time.

M stakes · generic model fails · individual segment

passes

Voice system for a visible founder

Reputation is the asset. Generic copy can imitate a tone, not a person’s history and never-list.

M stakes · generic model fails · individual segment

fails one

AI summary of public industry news

Interesting, but the buyer loses nothing when the summary is late or generic.

Low stakes · reject before build

fails two

Generic meeting note cleanup

A prompt already produces the useful answer. There is no encoded judgment to sell.

Generic model works · reject before build

IDEA FILTER SHEET

Buyer: [name the individual]
Loss: [what disappears if they get it wrong]
Private context: [what a generic prompt cannot know]
Lab gap: [why enterprise tools do not serve this buyer]

PASS only when all four lines are specific.