The air purifier lesson for AI products

TL;DR: The useful AI product lesson from “Air Purifiers as White Noise Machines” is that users often adopt tools for secondary jobs, not the job printed on the box.

What does an air purifier have to do with AI?

The primary source here is the Hacker News item titled “Air Purifiers as White Noise Machines.” That is all the supplied material gives us. No product test. No decibel chart. No medical claim. Just a small framing move: a device sold for air quality can also be valued because it makes a steady sound.

That tiny reframing is more useful than it looks.

A lot of products win because of the thing they do second. The air purifier cleans air, at least in theory and depending on the unit, room, filter, placement, and maintenance. But a person might keep it on every night because it masks street noise. The side effect becomes the retained value.

AI products have the same pattern. A meeting summarizer is sold as “never miss action items.” In practice, the hidden value may be that a manager can skip rewatching recordings. A coding agent is sold as “write code faster.” The sticky value may be that it gives a junior developer a safe first draft to argue with. A chatbot on a support site is sold as deflection. The actual internal win may be finding gaps in the docs because users keep asking the same missing question.

The lesson is not “position everything as something else.” That gets cute fast. The lesson is to watch what users keep using after the novelty is gone.

How do you find the hidden job in an AI workflow?

Ask for the moment, not the feature.

“What do you use this for?” gets you the brochure answer. “When did you last open it?” gets you closer. “What would have been annoying if it had not existed?” is better. With AI tools, I like looking for repeated low-drama usage. Not the demo. Not the viral clip. The third Tuesday use case.

That is where the real product shape shows up.

If people use your AI research assistant mostly to turn messy notes into cleaner notes, stop pretending the main value is deep autonomous research. If a sales team uses your agent to prep call briefs but does not trust it to send emails, build around prep. If marketers use a content system mainly to create outlines and internal briefs, the product is not “one-click publishing.” It is structured thinking under time pressure.

This is also where many AI products overreach. They pitch the full replacement and then discover users wanted a calmer intermediate step. Draft this. Compare these. Pull the likely answer with citations. Flag the odd item. Make the next human action easier.

That can look smaller. It often sells better.

Why this matters more with AI than normal software

AI has a trust gradient. Users may be happy to let a model propose, summarize, sort, or critique. They may hesitate when it acts, sends, purchases, deletes, publishes, or speaks to customers. The hidden job often lives at the trust boundary.

That means product teams should separate three things: what the model can technically do, what the user will permit, and what the organization will reward. Those are not the same.

A support agent may be capable of refunding an order, but the business may only trust it to draft the refund note. A legal assistant may summarize a contract well enough to save time, while nobody wants it making the call. A coding tool may generate a working patch, but its most valuable role may be explaining the scary legacy file before a human edits it.

The Hacker News framing is useful because it is mundane. No grand theory. Just a reminder that a thing can be bought for one promise and kept for another.

For builders, I would test the “white noise” value of any AI product before adding another big feature. Watch five real sessions. Find the quiet repeated behavior. Ask what users would miss if you removed it. Then make that path faster, safer, and more obvious. The catch most teams miss: the hidden job may be less impressive than the pitch, but it is usually closer to revenue, retention, and daily habit.