How this shop decides what to build, and how it reaches anyone

Intercept demand with free products that select for the intent your paid product serves. Borrow only to jumpstart. Own the list as a byproduct, never as a project. Keep the paid door inside the product.

Pick the audience by whether you can name three products for it.

The obvious reading of the ledger was that asynchronous channels work and channels needing a person present do not. Two rows broke it, from opposite directions. Filotto sent a chapter every night with no human in the loop at all, to a list of one, and produced nothing in a week. The best-converting channel I have ever had was entirely presence-bound: conversations, meetups, direct messages, no pitch energy required.

So cheapness buys nothing on its own. What both rows share is worse to hear:

Every channel that produced anything spent someone else's hours. Every channel that failed spent mine.

A store's search index spends the store's hours. A referral spends the referrer's. A community spends attention someone else already assembled. Announcements, check-in emails, newsletters and live sessions all spend mine, and mine are the one input I have not had a contiguous block of since May.

Intercept. The demand already exists and you put yourself in its path. Store search, web search, ecosystem registries. Requires no audience and no reputation, which is why it comes first.

Borrow. Someone else has assembled attention and lets you touch it. Communities, referrals, other people's newsletters. Requires standing. Trust transfers where cold broadcast does not.

Own. An audience you assembled yourself. Requires that the audience already exists, which is the part that gets skipped.

The order is a dependency, so it cannot be rearranged. Intercept and borrow are the inflows and own is the reservoir they fill. Build the reservoir first and you get a well-engineered thing attached to nobody, which I have now done twice.

• Nothing enters the portfolio until I can name four things: its audience, its channel, its posture, and whose hours that channel spends. If the only posture I can name is own, against an audience of zero, I do not build it.
• Owned channels are byproducts, never projects. Any list that needs a recurring act to grow is a job, and I cannot staff a job.
• Monetization is a rung inside the product, never a campaign aimed at a list. A rung costs me nothing after I build it. A campaign costs me an evening every time.
• Free products are the main lane here. For most people content is the cheap way to build an audience and products are the expensive way. I have that backwards: a product costs a review cycle, an essay costs the evenings I do not have.
• Every free asset gets a threshold and a date, both set before it ships. Revenue cannot be the kill signal in a free lane, and left to myself I do not kill things. Filotto ran nightly for weeks after its only reader stopped opening it.

I am building assets before businesses, so the free lane is scored on how many people it intercepts rather than what it earns. The number to beat is 29 downloads a month, which is the best rate anything here has managed.

Deferring revenue does not relax the clause about intent. It leaves that clause doing the job alone. While there is nothing to sell, nothing corrects an audience assembled on the wrong intent, and the error compounds for months before anything reveals it.

The whole model rests on an assumption I have never measured: that one asset can intercept meaningfully more than 29 downloads a month. A single listing pass on Pet Med Reminder, an app about giving the dog its tablet, measures that and nothing else. The pass is written and sitting in App Store review. The number arrives when the update does.

Until it does, I am not scaling the model. That measurement is the first objective of the current cycle, which is how a pet medication app came to be the most strategically important thing I own this fortnight.

I checked six candidate audiences against one question: is this job already served? The method was to search the relevant directory for the job in a user's own words and count the shipped competitors. Zero or one is open. Two or three is contested. Four or more is closed. The search takes about a minute.

Five came back closed, several by four or more shipped products, one by a feature built into the platform itself. Which produced the finding worth more than the rest of the research together:

In 2026, every job you can reach by browsing a directory has already been built, because building is cheap for everyone now and not only for you. The durable openings are where the barrier is something other than build cost: proprietary experience, credibility, or a specificity too narrow for a generalist to bother with.

So I stopped browsing stores for gaps. The audience left on my list is the one whose barrier is standing rather than build cost: engineering leaders moving their teams onto an AI-native process, which is work I have done. Whether that is a product or only something I know is the third question in this cycle.