Automatic memory: the category nobody named yet

The phrase does not show up in any product tagline. Type it into a search bar and you get results about computer RAM, or about automatic speech recognition with the word “memory” nearby. The category has no Wikipedia page. No one has put it on a conference slide. But the thing it describes is everywhere, and you have almost certainly used one this week.

When Spotify generates your Wrapped every December, that is automatic memory. When Apple’s Photos app quietly builds a “Trip to Kyoto” video from the pictures you took in October, that is automatic memory. When GitHub draws your contribution graph, when Strava produces a year-in-review of your runs, when Swarm reminds you that a year ago today you checked in at the café on the corner, when Google Photos surfaces “this day, three years ago” at the top of your feed, every one of these is automatic memory.

The pattern is identical. A service you used for some other reason (listening, photographing, running, coding) turns the trace you left behind into something you can feel. You did not ask for it. You did not write it. You did not even know it was being kept.

And yet none of these products call what they are doing “automatic memory.” They call it Wrapped. They call it Memories. They call it Year in Review. Each one treats the feature as a marketing flourish attached to a different product. The underlying activity, assembling what you did into something you can re-encounter later, has never been promoted to the status of a category.

Why category names matter

A category name is not just a label. It is an organizing force. Once a name takes hold, the scattered instances of the thing realize they belong together, and both users and builders start to orient around the newly legible shape.

“Social graph” is the canonical example. Facebook did not invent the underlying idea, the notion that the web of relationships between people could be modeled as a graph. Social scientists had worked in that frame for decades. But when the phrase moved into product vocabulary, a whole industry arranged itself around it. Products started claiming to map it, extend it, or compete with it. Without the name, there was a pile of features. With the name, there was a category.

“Second brain” did the same for personal knowledge management. The practice of taking structured notes to augment biological memory predated Tiago Forte by centuries. But the phrase turned a private discipline into a shared vocabulary, and a small mountain of apps, courses, and communities grew up around the name within a few years of its coining.

Automatic memory is in the stage that comes before naming. The instances exist. The builders shipped them for other reasons. The users love them. Nobody has said out loud that they are the same thing.

What the name points to

An automatic memory is any product that takes a trace you left while doing something else and returns it to you later as something that feels like a memory rather than a record.

Two halves. The trace is automatic, meaning you did not set out to log anything. You were listening to music, walking with your phone, closing tickets, going to the dentist. You left a footprint because the tools you used keep footprints. The return is also automatic, meaning the service decides when and how to hand the trace back. The composition is not yours. You did not query for it. It arrives.

The shape of what arrives is the distinguishing feature. A Slack export is not automatic memory, because an export is a file, not a memory. A GitHub contribution graph is not obviously automatic memory either, because the squares have no texture. But when Strava sends you a year-end story with miles, hours, elevation gained, and your longest run, shaped into a narrative you scroll through in five minutes, the output crosses into memory territory. Something you did all year, mostly forgotten, becomes feelable again.

The line is where the raw data stops being information and starts being experience. Most products that produce automatic memory got there by accident, not design.

The accident

Almost every automatic-memory feature in the wild was built as a retention tool. Spotify Wrapped exists because sharing it on social media drives new installs. Apple Memories exists to keep people inside the Photos app. Strava’s year in review exists to make runners reopen the app in December. GitHub’s contribution graph exists because visible progress increases engagement.

The features succeed because they trigger a feeling the marketing team did not fully anticipate. Users reopen the app and find themselves staring at evidence of their own year. The dentist appointment nobody remembered. The song played fifty-three times in April and never again. The run they were sure they only did once, which turned out to be a streak of eleven weeks.

The feeling is not “here is my data.” It is “oh, that was me.” A small recognition, not a dashboard.

Because the feature was designed as a retention hook, it arrives once a year. Because it was designed as a retention hook, it is stylized and shareable, not browsable. Because it was designed as a retention hook, it ends when the season ends, and the rest of the year you are back to a product that does not remember anything on your behalf.

This is the shape of a category whose instances are all side effects. The feeling is real. The category is real. Nobody has built a product around it as the main event.

What automatic memory is not

To name a category is also to draw its edges. Several adjacent categories get confused for this one, and the confusions are instructive.

It is not lifelogging. Lifelogging was about capture: wear a camera, record everything, store the stream. Capture without composition is not memory, it is storage overhead, which is why the lifelogging movement collapsed. Automatic memory takes existing digital traces and shapes them into something usable; the capture problem is already solved by the apps you used for other reasons.

It is not quantified self. Quantified self is the discipline of collecting your own metrics for analysis and optimization. The artifact is a chart. Automatic memory is not trying to optimize anything. Its artifact is not a chart but a scene, and it exists so you can re-encounter the day, not improve the next one.

It is not a second brain. A second brain is a writing practice. You create the notes, curate them, and link them. The artifact is the index you maintain. Automatic memory requires no writing and no curation. The artifact is generated from activity you were already producing.

It is not journaling. A journal is something you keep by composing it. Automatic memory does not ask you to compose anything. We wrote separately about the difference between these two shapes of diary.

What each of these adjacent categories has in common is that they ask something of you: effort, attention, discipline, a habit. Automatic memory’s defining property is that it asks nothing. The trace was produced anyway. The return happens anyway. You are the recipient, not the author.

The category sits underneath automatic journaling

Some of our own writing on this blog has been about automatic journaling as a practice and the automatic journal as an object. The category of automatic memory contains both of these, and more.

An automatic journal is one way to deliver automatic memory. It is the daily form. The pipeline runs overnight, the diary entry appears, and over months you accumulate a continuous record. That is a specific instance of the broader category.

Spotify Wrapped is another instance, shaped differently. It delivers once a year. It looks back over twelve months, not one day, and the medium is animated slides, not prose. The category is the same. The form is different.

Apple Memories is yet another, delivered on an irregular rhythm. A holiday triggers it. An anniversary. The trip you took last September. The pipeline runs quietly in the background, and occasionally a new movie appears at the top of the Photos app without announcement.

Once you have the category name, you can see the design space. A daily form. A weekly form. A monthly form. An annual form (Wrapped, Year in Review). A location-triggered form (walking past the café where something happened). A relationship-triggered form (a person reappears in your day and the service surfaces the last time they did). Each of these is a different delivery schedule for the same underlying primitive.

Most of the design space is empty. That is usually what it looks like when a category has not been named yet. The instances exist in isolation, each solving the same underlying problem for a different trigger, and nobody has noticed they are variations of one idea.

Why unnamed categories are under-built

A feature inside a product gets exactly as much investment as the product’s primary metric allows. Wrapped will never be more than a marketing moment for Spotify, because Spotify is a music subscription service, and its primary job is to keep you streaming. Memories will never be a standalone artifact for Apple, because Apple needs you inside Photos. Contribution graphs will never become a personal diary for GitHub, because GitHub is a code host.

Every automatic-memory feature inside a bigger product will be capped at whatever the bigger product needs it to be. That is fine. Those features are good at what they do. But if automatic memory is a real category, and we think it is, then it deserves products that treat it as the main event, not as a retention hook.

Names unlock that. Once a category is legible, investment follows. Standards emerge. Users form expectations they carry across products. The thing becomes discussable at a level above any individual implementation.

The bet

Our bet with deariary is that automatic memory is worth treating as the product, not as a feature. A daily-form instance of the category, running every night, building an archive you can re-encounter any morning you choose. It is the form of automatic memory we found most useful, which is why we built it first. Other forms will make sense too. We will build some of them. We hope other people build the rest.

This post, more than most on this blog, is trying to do something rather than describe something. We think the category exists and is useful, and we are going to keep using the phrase until it either catches on or fails. Both outcomes are fine. What we want to avoid is the state it is in now, where every product that ships an automatic-memory feature calls it by a different marketing word and treats it as a side dish.

The thing you felt the last time a Wrapped video caught you off guard, the thing you felt when Apple handed you a video of a trip you had half-forgotten, the thing you felt when your contribution graph reminded you how much of last March you spent heads-down: that is automatic memory. It already exists in pieces. It does not yet exist as a category.

We think it should, so we are using the name.