Buy the answer instead of arguing about it

We had a genuine disagreement about how to produce visual content for a video series. A ten-minute piece with a scene change every five seconds needs roughly 119 shots. Three approaches were plausible.

• Full generative motion for every shot — the most impressive, and at metered per-second pricing, roughly $67 per video.
• Generated stills with programmatic pan and zoom — about $10 per video.
• A hybrid: motion where it earns its cost, stills elsewhere.

There were credible arguments in every direction. Motion holds attention. Stills with a slow push read as intentional camera work, especially in a flat illustrated style. The hybrid sounds like the obvious compromise and might be the worst of both.

The trap of deciding by discussion

This is the kind of question a team can argue about for a week. Everyone has an opinion, all the opinions are defensible, and none of them are evidence.

Worse, the argument usually resolves by seniority or stamina rather than by being right — and then the decision calcifies, because revisiting it means reopening a fight.

The recurring cost of getting it wrong here was material: at weekly publication, the expensive path runs about $269 a month, and the cheap path about $40. Over a year, that gap is the difference between a viable and unviable production model.

So we built all three

We added a mode flag to the render pipeline and produced the first piece three times — full motion, stills, hybrid. Total cost, roughly $95. One afternoon.

Then we watched them. Not evaluated the architecture, not reasoned about retention research — watched the three outputs and asked which one we'd rather have published.

$95 to replace an opinion with an artifact is not a cost. It's the cheapest research you will ever buy.

When this is the right move

The tactic works when three conditions hold, and it's worth checking them explicitly:

• The experiment is cheap relative to the recurring cost of being wrong. Ours was roughly a third of one month's difference.
• The comparison is legible. You can look at all three results and form a judgment without a study.
• The result is durable. This decision governs every future video, so the answer keeps paying.

It's the wrong move when the experiment costs more than the decision is worth, when you can't tell the outputs apart without formal measurement, or when the answer will be obsolete next quarter.

The part people skip

Build the comparison into the tool, not around it. Because the mode became a flag on the pipeline rather than three hacked-together branches, re-running the comparison when models and prices change costs another afternoon rather than another project.

Generative pricing moves constantly. A decision made on this quarter's numbers deserves a cheap way to be revisited, and the version of this experiment that lives in the codebase gets revisited. The version that lived in a spreadsheet does not.