Personal AI is quietly replacing the apps you pay for
Two high-school kids photographed their lunch and built a thirty-million-dollar company.
The app is Cal AI. You take a picture of your food, it counts the calories, and that is the whole product. Two teenagers, Zach Yadegari and Henry Langmack, launched it, passed 15 million downloads and over $30 million in annual revenue in under two years, then sold it to MyFitnessPal. Tens of millions of dollars, from a photo of a sandwich.
Now the part that should stop you. The thing inside Cal AI that looks at your plate and knows it is chicken and rice runs on image models from Anthropic and OpenAI. The same models sitting in the Claude or ChatGPT app on your phone right now. This is what personal AI quietly hands you: the engine behind a company that just sold for a fortune.
So I have been stuck on an uncomfortable question. If the intelligence is something I already pay around fifteen dollars a month to use, why am I also paying for the app wrapped around it?
That question is the whole future of personal AI. The answer is stranger and more useful than the hype makes it sound.
What people are building instead of buying
Picture the version where you skip the app. You photograph the meal, send it into a flow you wired up once, and the AI reads the plate, logs it, and pings you on WhatsApp or drops it into a table you can open later. No subscription. No download. Just a bit of plumbing that runs whenever you need it.
A year ago that sounded like a hobby for engineers. It is starting to look like a habit.
The early signs are easy to miss if you are not watching for them. Claude now keeps live artifacts you can come back to, little tools that stick around instead of vanishing when the chat closes. Connectors let your AI reach into your email, your calendar, your files. And on January 30, 2026, Anthropic shipped plugins that let non-developers automate workflows that used to need five to ten separate SaaS subscriptions.
The market noticed fast. In February 2026, roughly $285 billion in value vanished from software stocks in a single trading session. People started calling it the SaaSpocalypse. It is not a fringe theory either. Gartner projects that 35% of point-product software tools will be replaced or absorbed into AI agent ecosystems by 2030.
Here is the principle underneath all of it. The app was never the intelligence. The intelligence is becoming a utility, like electricity, and you do not buy a new company every time you want to boil a kettle.
Now let me argue with myself
This is where the clean version of the story falls apart, and I would rather break it myself than let you find the crack later.
If Cal AI is "just" the image model I already have, why did MyFitnessPal pay tens of millions for it instead of building a flow in an afternoon? Because the AI was never the moat.
It grew through creator content on TikTok and X, not through being clever. That is distribution, and distribution is hard. Over 30% of people kept using it, which means it built a habit, and a homemade flow has no habit loop. It worked in one tap, while my flow needs setup, breaks when a connector changes, and has exactly one person on the support team, me. And MyFitnessPal owns a database of 20 million foods and meals from more than 380 restaurant chains. My flow guesses.
The value was never the calorie number. It was the friction the app removed and the trust it earned. That part does not collapse when the model gets cheap.
Then there is the device, the piece everyone gets most excited about. OpenAI bought Jony Ive's hardware company io for about $6.4 billion and confirmed a screenless, voice-first device for the second half of 2026, though a court filing suggests it could slip to early 2027. The Kindle-instead-of-iPad logic is tempting. A calmer, single-purpose thing built only to serve your AI.
But we have seen this movie, and it ended badly. Humane sold its AI Pin as the future of personal computing, shut the servers down on February 28, 2025, and sold to HP after shipping fewer than 10,000 units. The Rabbit R1 shipped around 100,000 units but had roughly 5,000 daily active users. The lesson the whole industry wrote down was blunt: a standalone AI device has to beat the smartphone, not just beat nothing. The hardware actually winning is the boring kind, AI tucked inside glasses people already wanted, where Meta's Ray-Bans took 75 to 80% of the market.
So the Kindle comparison cuts both ways. The Kindle won because e-ink fixed one specific, real pain. Most single-purpose gadgets since then are sitting in a drawer.
So what actually changes?
Split the thing into two layers and it gets clear.
The commodity layer is the AI doing the obvious work: reading the receipt, drafting the email, recognizing the food. That layer falls to almost nothing and moves inside your personal AI. The thin apps that were only a login screen wrapped around that one trick lose their reason to exist.
The value layer is everything that made you keep coming back: the frictionless habit, the trusted data, the distribution that put it in front of you. That layer survives, and people keep paying for it.
The apps most exposed are the ones that were a single feature with a subscription attached. The ones that survive own data, a habit, or a community that a flow cannot fake.
And the money does not disappear. It moves up. It goes to whoever owns the aggregation, the Anthropics and OpenAIs whose models everyone now rents, and to the connectors in the middle. You save the subscription. The platform keeps the margin the app used to take.
Who actually gets paid in this world
Two new jobs appear in the gap.
One is that ambient device, for people who want AI without another screen in their face. Real opportunity, but a bet, and Humane is the reason I would not wager the house on the gadget winning first.
The other has no name yet. Call it a personal AI architect. Right now there are people charging real money to wire up automations, but almost all of that work is for businesses. The version that sets up your life, the quiet flows that run your week, barely exists. That gap is the opening.
Here is the catch, and it is the part that keeps me up. The same platforms racing to make this possible are racing to make it trivial. The day building a flow becomes a template you tap once, the expert who charged to build it is as replaceable as the app was. So the durable thing was never the building. It is taste. Knowing what is worth automating and what to leave alone. The judgment, and the trust it takes for someone to hand you their inbox and their data.
Strip all of it back and people are not chasing flows or devices. They want to stop paying rent on their own routines. A subscription you forgot you signed up for is a small monthly reminder that someone else owns a piece of how you live. Personal AI, at its best, hands that piece back to you.
Cal AI did not lose, by the way. It sold to the incumbent it was disrupting, and the value moved exactly where this whole argument says it would, up and into the company that owned the data and the habit.
Which is why the goal was never to build the next Cal AI. It is to help one person build the small, slightly ugly, perfect-for-them flow that means they never need to buy one. If you want to start, skip the device and the grand system. Photograph one thing you do every week, send it to your AI, and tell it where to put the answer. That is the whole future, and it fits in an afternoon.
Common questions
Will AI replace the apps I pay for?
Some of them, yes, but not the way the headlines say. The apps that were just a thin wrapper around AI lose their reason to exist, because you already rent that same intelligence. The ones that own your data, your habit, or a real community tend to survive.
What is personal AI?
It is using the AI tools you already pay for, plus connectors to your email, files, and apps, to build small workflows that do a specific job for you. Instead of buying a separate app for each task, you assemble your own flow once and reuse it.
Do I need a separate AI device for this?
Not today. The screen-free AI devices are coming, but the early ones like the Humane AI Pin failed because they tried to beat the phone and could not. For now, your personal AI lives perfectly well inside the phone and laptop you already own.