What I stopped paying for the week ChatGPT learned to remember

The morning persistent memory landed inside the model I already pay for, I opened my phone and cancelled seven subscriptions before breakfast. I am writing this down because nobody in the wellness category will tell you which of their categories just collapsed, and you have a right to know.

The seven I cancelled

None of these apps were bad. Several were beautifully made. They all had the same problem: they were thin wrappers around a model that has now learned to remember me without paying them for the privilege.

• An AI symptom checker — the model now holds my medication list, my last three lab panels, and my chronic conditions across every thread. The app was charging me eight pounds a month to re-paste context I had already typed.
• A nutrition coach — I had been logging meals for it to "learn" my patterns. The model already had eight months of grocery receipts from my email and had inferred more than the coach ever surfaced.
• A sleep journal — replaced by a single saved instruction: "At 7am, ask me one sentence about last night."
• A women's cycle tracker — not relevant to me, but my partner cancelled hers the same week, for the same reason.
• A "habit AI" — turns out a habit is just a sentence the model now reliably remembers.
• A wearable's premium tier — I kept the device, dropped the subscription. The CSV export and a chat thread does what their dashboard did, with more honesty about uncertainty.
• A meditation app's "AI coach" — meditation is still good. The chat about meditation is now free.

Combined: roughly €78 a month. Just under a thousand euro a year, gone the morning a feature rolled out that I did not have to choose, did not have to install, and did not have to learn.

The one I doubled down on

I extended my subscription to the chat tool itself to the annual plan, and started paying for a second one — Claude — because for medical documents it still refuses to invent. Two subscriptions, less than half what I was spending on the satellite apps that orbited the same models, and I now hold all my context in two threads I will still own next year.

This is the part the category does not want to say out loud: the foundation models are quietly eating their own application layer. Anything that was just persistence plus a friendlier UI on top of a chat model is on borrowed time. What survives is what models genuinely cannot do alone.

What still earns its line on the bank statement

A small, honest list. I kept the wearable hardware, because skin contact and a heart cannot be replaced by software. I kept a private notes app — Obsidian, markdown, on my disk — because I want a copy of my own ledger that does not need an internet connection or a corporate roadmap to exist. I kept my clinician, more on that in a minute. And I kept a quiet membership to one community of people who are trying to do this carefully, because no model has yet learned to be embarrassed in front of friends.

Everything else is now a sentence I say to a chat thread. The thread remembers. That is the whole shift.

What this means if you are not me

You probably are not running an experiment on yourself for a living, so let me be careful with the recommendation.

I am not telling you to cancel anything. I am telling you to audit. For each AI-flavoured health app on your phone, ask one question: "What does this do that a chat thread with memory cannot?" If the honest answer is "it remembers me, it is friendlier than the chat box, it has a pretty graph" — that app is now optional. If the honest answer is "it touches my body, it owns data I cannot export, it provides a human on the other end" — keep it.

This is not anti-app. It is anti-paying-twice. You are already paying for the model. Anything sitting on top of it that does not add a layer the model cannot reach is, today, charging you rent for a room you have already bought.

The thing AI is not coming for

I want to be very clear about the part that did not change.

The week persistent memory rolled out, I still had a forty-minute call with my GP about a marker on my last panel that had drifted. The model had flagged it weeks earlier — that part is real, and useful, and I would not have noticed without it. But the conversation about what to do, about my risk tolerance, about whether to retest or to act, about what my mother died of and what that means for me — that conversation belongs to a person who can be sued, who has skin in the game, who knows the difference between a confident sentence and a careful one because a career hangs on it.

The chat thread is the best preparation for that conversation I have ever had. It is not a substitute for it. The week the model learned to remember, the value of my clinician did not go down. It went up. I walked in with better questions, and we got further in forty minutes than we used to get in two visits.

Cancel the wrappers. Keep the humans. The membership economy of personal health is about to look very different, and the people who notice early will spend less and learn more.

—

This is the first essay in The Journal. One a month, long enough to be worth the read, written by a person, not a content engine. The next one is about the wearable that does not know what kind of week you had.