Storytellers Get Paid First

A new field pays for stories before it pays for work. Nobody can measure quality yet, so the money attaches to the story.

Most builders in my field believe the story is what you reach for when the work is weak. I believe the story is a component of the system, with its own failure modes, its own tests, and its own owner. The rest of this essay is the case for building it like one.

The market funds the only signal it can read

The complaint from builders is accurate as far as it goes. Investors back the best pitch instead of the best system, and attention goes to whoever talks loudest. The complaint rests on one assumption, which is that a buyer can tell the right people from the wrong ones.

Pricing quality needs a way to measure it, and a new field hands you none. The benchmarks are games, because a number everybody trains against stops measuring anything. The track records run too short to contain a real failure. A buyer holding two proposals cannot separate a system that works from a demo that survived one afternoon.

Quality nobody can measure is quality nobody can price. One signal still travels from a builder to a buyer without a measurement attached, and that signal is the story. The money follows it, which looks like a preference for talkers. Blaming buyers for funding storytellers is blaming them for using the only instrument they hold.

The demo takes days and the product takes months

I watch this from the inside. My job starts after the demo. The demo takes days and collects the applause. The product takes months, and the months go into three failures.

The retrieval misses on the query nobody thought to type, and finding that query costs more than building the retrieval did. The eval passes while everybody watches and fails after they stop looking. Whoever is still in the code pays for that one. Nobody wrote the edge cases down, so a user meets each one first, and the repair costs more than the code it touches.

None of that shows up in a pitch, which is why the pitch works. A pitch is a demo with a narrator.

Here is the asymmetry that changed my mind. The demo walks into the room by itself. The months behind it reach a buyer only if somebody carries them, and the carrier is a story. Skip the job and the months stay invisible, which prices them at zero. The market prices what it can see, and I control what it sees.

The correction arrives on the market's schedule

Buyers do learn to measure. The learning takes cycles of getting burned, and then somebody finds the duct tape. A public company bought other companies and cut their research budgets. Investors paid for the cut costs for years, and then the pipeline came up empty.

The distance between the loud first effect and the quiet later one is the effect horizon. The correction is real, and it arrives on a schedule the builder does not set. Day two hundred of a build is dull and nobody is watching, and it is also the day a buyer has nothing to read. Plenty of good builders do not survive the wait.

Builders draw a different lesson and decide storytelling is beneath the work. They inherit that lesson rather than test it. It is a borrowed belief, untested by its owner, and the price of keeping it is readers.

I resented that character for a long time, the one who cannot build and gets funded anyway. Then I priced the wait, and the resentment stopped.

Constraints separate a story from hype

So I decided to get good at the story, under constraints that keep it checkable. Four of them, and each one is a test a reader can run.

A real story makes one claim, and the claim holds when somebody in the field pushes on it. A story that makes many claims makes none, because the reader keeps whichever one is easiest to believe.

It shows the before and the after, including the run where the edge broke. A reader who can check that against the running system comes back.

It names the cost: what the thing takes to run, and the places it still fails. A cost named before the trial is information. A cost the buyer finds during the trial is a reason to stop.

The story makes no prediction. A story that stops at what happened survives the next release. The next release voids a story that reached into the year ahead. Hype carries none of these constraints, which is what makes it easy to sell, and what makes it die on contact with a bad week.

The strongest objection, and where it holds

The best case against this essay: practice at storytelling is practice at seduction, and the practiced drift toward the incentives of the loud. The drift is real. I have watched a team polish the narrative of a system while its eval set went stale. The narrative won the quarter.

The constraints are the answer, and one rule sits under them. Where I cannot say the true thing, I say nothing. Silence costs a news cycle. A false story costs the audience that made the true ones worth telling. If a story of mine fails its own constraints in the field, send me the failing run. I will retract it in the channel I told it in.

Tell it only if a bad day leaves it true

One test decides whether the story is ready. If the system had a bad day, is the story still true? If it holds, tell it.

If a bad day makes the story false, the story is not ready, because the system is not ready. The work in front of you is the system, and the story waits for it.

I do not expect the measurement problem to be solved soon. So the job is to become a builder a buyer can hear.

Take the last thing you shipped and give it a bad day on purpose. Write the story that is still true afterward. The sentence you had to delete names the next thing to build.