A Short History of Sherlocking
In 2002, Apple shipped Sherlock 3, a search tool for Mac OS X. It happened to replicate, almost feature by feature, a beloved third-party app called Watson — built by a small developer, Karelia Software, which had done the hard work of figuring out what users actually wanted. Watson didn’t survive, and the episode gave the industry a verb: to be sherlocked is to watch the platform owner absorb your product into the operating system, free of charge.
It became a ritual. f.lux was sherlocked by Night Shift. Duet Display by Sidecar. An entire generation of flashlight apps by a single toggle in Control Center. More recently, a good part of the password manager market woke up to find Apple shipping a dedicated Passwords app. Every June, developers watch the WWDC keynote with a very specific kind of anxiety — hoping their roadmap isn’t on the slides.
Why It’s Bad
The problem isn’t that platforms improve — of course they should. The problem is the asymmetry. The platform owner sees the sales data, controls the distribution channel, collects a commission on your revenue while you validate the market for them, and then decides whether your category is worth taking. The developer carries all the discovery risk; the platform captures the reward at near-zero marginal cost.
The rational response is a chilling effect: don’t build anything a platform could trivially absorb. Build deep, build weird, build things with defensible complexity. That advice has aged well — better than anyone expected, actually — because the absorption machine just changed owners.
Enter the App That Builds Apps
Tools like Lovable and Replit have turned “describe what you want” into a working application with a database, authentication and hosting. Not a prototype — the actual thing. The demos tend to oversell the polish, but the direction is unambiguous.
And it’s already going one layer deeper: apps that generate other apps within themselves. Meta is experimenting with exactly this — Pocket, quietly launched a few days ago, lets users prompt playable mini-games (“gizmos”) into existence and share them in a feed, like user-generated content where the content is software. It’s a toy today. So was the App Store, briefly.
If I Know What I Want, I Can Just Ask for It
This is the part that matters for the end user. Think about what a huge share of the SaaS market actually is underneath the branding: a form, a table, a filter, a chart, a notification. CRUD with a nice coat of paint. Habit trackers, invoice generators, booking pages, inventory sheets, link-in-bio pages — useful, yes, but structurally simple.
If I know what I want, I can now just ask for it. A tool that fits my exact workflow, my columns, my edge cases — instead of paying a monthly subscription for someone else’s compromise and a settings page full of features I’ll never touch.
And if I’m more technical, I don’t even need the friendly wrapper — any of the agentic coding harnesses (Claude Code and its many cousins) will happily build and maintain the whole thing from a conversation. The gap between “I have a spreadsheet problem” and “I have software that solves it” has collapsed from months and a funding round to an afternoon and a prompt.
The Great Cannibalization
Here is where sherlocking returns, at a different scale. When a major platform ships this capability natively — and Meta’s Pocket suggests they’re already rehearsing — it won’t sherlock an app. It will sherlock the category. Every niche SaaS that is pure CRUD becomes a prompt away from irrelevance.
The irony is that the platforms will be cannibalizing their own revenues to do it. Those thousands of small SaaS businesses pay for cloud infrastructure, app store commissions and advertising — much of it flowing to the very companies now building their replacement. They’ll do it anyway, for the same reason Apple never hesitated: better to eat your own margins than to let someone else do the eating.
What Survives
I don’t think this is the end of software businesses — it’s a repricing of what was defensible all along.
There is still value in bespoke software: systems with real domain depth, regulatory weight, integrations that took years of relationship-building. There is still value in knowing what to build — PRDs, requirements work, consultancy. Ironically, the scarce skill was never writing the code; it was specifying the problem. (Anyone who has watched a client discover what they actually wanted three sprints in already knew this.)
But the simple SaaS — the $29/month wrapper around a database table — is dead. It may take a few years for the funerals, as these things always do, but the economics are settled. The edge migrates to those who were structural all along: the scalers who run the compute, the gatekeepers who own distribution, the model providers who own the intelligence, and hard tech that can’t be prompted into existence.
A New Abstraction Layer
For the end user, none of this will feel like an industry collapsing. It will feel like software finally behaving the way documents always did. Nobody licenses “a spreadsheet about my finances” — you open a blank sheet and shape it. Software is becoming that: something you shape, not something you shop for.
We spent decades treating applications as products — fixed artifacts, designed for the average user, sold in units or subscriptions. That was never a law of nature; it was a limitation of how expensive software was to produce. Assembly gave way to compilers, compilers to frameworks, frameworks to no-code — and each layer let more people shape computation with less ceremony. Natural language is simply the next layer, and the last one most people will ever need.
We just found a new way to shape information.