The Big Company Did the Math, Then Walked Away

Last weekend, late at night, I scrolled past another one of those posts: "Solo dev goes it alone with AI, hits $7k a month in three months." My first reaction wasn't envy. I zoomed in on the screenshot of the product he'd built and stared at it for a bit, quietly estimating: if Feishu or WeChat wanted to build this, how many people and how many days would it take to ship? Once I finished the estimate I felt a little deflated. If the answer is "three people, one week," that business is basically on thin ice.

That reflexive estimate comes from an old term I've been watching for years: sherlocking.

Here's where it comes from. In 2002, Apple's Mac operating system had a little search tool called Sherlock. Around then a small company called Karelia built a better search tool, Watson, which could pull up weather, stock quotes and phone numbers, sold for a few tens of dollars a copy, and did fine. That year Apple shipped Sherlock 3, which copied nearly all of Watson's features, built them into the OS, and gave them away for free. Watson's business was gone overnight. Ever since, "app X got Sherlocked" has been the standard phrase for a small developer's hard work getting lifted wholesale into the platform's own system, livelihood cut off.

There's a long list of these. F.lux, which adjusts screen color temperature: Apple built Night Shift into the OS. Duet and Luna Display, which turn an iPad into a second monitor: Apple built Sidecar. A pile of screen-time management tools: Apple just put "Screen Time" right into Settings. The same thing happens every year inside China's platform ecosystems too, it just doesn't have a catchy name there.

Why I keep an eye on this term: I've been doing tech for about ten years, and on the side I maintain two small projects, one a knowledge base for collecting material, the other a set of scripts that helps me write. Every time I think about adding a direction to them, or catch myself wondering whether to actually turn one into a product, the first question I can't get around is this: has a big company looked at this direction, and if so, why didn't they build it?

First I have to be clear about how the decision works on the big company's side, or it's easy to feel like they're coming after you personally.

Whether a big company adds some small feature to its system comes down to basically two things. One: how many people actually want it. The product folks have a term, addressable market, which plainly means how many users this feature would reach, whether it shows up as a line on the earnings report. Two: how expensive it is to build, how many people and months, whether it competes for resources with the company's other work.

Run those two numbers and the conclusion falls out. A feature that enough people are loudly asking for and that isn't much trouble to build, they just add it, no extra reason needed, and they genuinely won't stop to check whether they've taken the bread off some small developer's table. At that scale, your little tool with a few thousand users rounds to zero in their data. Conversely, when a big company hasn't built something for a long time, it's rarely because they didn't see it. More often they did the math, decided the deal wasn't worth it, and walked away on purpose.

So "will a big company build this" isn't really a question about technology. It's a question about their ledger.

Flip that ledger around and you get the spot where people like you and me can stand.

The directions a big company did the math on and passed because it wasn't worth it usually look like a few things. The market is too small to feed a ten-person team, but big enough to feed one person. The work is too fragmented and messy to cover with one standard process, so you have to hand-hold one customer at a time. Or it is dirty and unglamorous: a heap of edge cases, grinding through requirements with users, and the return on investment just doesn't pencil out in a big company's model.

Those places are your mud pit. That you can stand there has little to do with your technical level; it's that you're willing to do, for that little money, the work a big company looks down on. I've written before about the export-file trap in the WeChat backend: the file one button spits out is actually three tables jammed together, and the error message only points at the surface symptom. No big company will ever build a tool for that, because maybe there is exactly one person on earth who needs it, which is me. And that is precisely why I can calmly write a script to fix it. Nobody is fighting me for it, and nobody is stopping me.

The cost lives in the same place. The flip side of nobody fighting you for it is nobody validating that the need is real. The ceiling is obviously low: you're in a mud pit, and the pit is only so big. Before you take this road, those are the entries you have to book first.

I said the big company's decision looks at two things, headcount and cost. What has changed in the last couple of years is the second one.

Some outlets have tried using today's AI coding tools to slap together the shell of a Monday-style project management app in an hour, iterating by just saying what they want in plain language. The cost of building the shell of a feature really has been pushed down a big notch by AI. That means the range of things a big company can just add has grown; mid-sized features it used to skip because they were annoying, it might actually build now. For someone whose whole footing is a single feature, the safe zone is shrinking.

But there's another side to this. Slapping together a shell is easy; the genuinely hard parts are ones AI can't carry: login and auth, permission boundaries, job scheduling, debugging production incidents, and grinding through requirement changes with customers over and over. Those are exactly the mud-pit jobs, slow and unsexy, and for now still work a person has to do. What AI lowered is the cost of copying one; what it didn't really lower is the cost of keeping it serviced for three years, and that second part is where the small player's real ground is.

Two signals worth watching going forward. One: has a big company tried building a version of your direction with AI. The other, and more important: after they built that version, did they actually maintain it, or did they ship it and leave it sitting there. Big companies using AI to quickly plow through a plot of land and then not bothering to tend it: that is only going to get more common. The plots that got plowed and then abandoned are still yours.