70,267.
That time, while organizing photos with AI, this number appeared in the report when the program finished. I stared at it for a while. The images I had accumulated had grown numerous enough that I needed another way to find them again.
This was the number of records produced by that pass. It reminded me of many moments when I had pressed the shutter, but I could not treat the number as more than seventy thousand distinct photographs or more than seventy thousand unique moments in my life.
An Accidental Digital Archaeology
When I started using Claude’s Cowork feature in January 2026, I initially wanted only to deal with long-postponed tasks on my computer: organizing files, consolidating subscriptions, and addressing various forms of technical debt. After working with AI to move some of that backlog forward, I reached the photos that had accumulated over many years.
I worked with AI to write a Python script that read photo metadata, including information such as time and location, organized the distribution by year, extracted geographic coordinates, and produced a JSON report. The process did not go smoothly. I had not set up the workflow properly, and my computer crashed several times.
When I finally saw the report, however, it felt like more than cleaning up files. Years and places I had not thought about for a long time came back into view. The data opened an entry point through which I could look back at my own time.
The Shape of Time’s Curve
In the distribution by year from that organizing pass, the number of images from 2018 stood out. Looking at it brought back the new encounters, career changes, and cross-border movement of that period. Cities were changing, work was changing, and travel and meetings were taking place in different locations.
Those associations matter to me. Still, the number of photos reflects the images that remained and were included in this particular pass, not a complete measure of how active my life was. Shooting habits, devices, duplicate copies, deletions, and decisions about what to keep can all affect the curve.
I cannot conclude that I did more in a particular year simply because more photos remain from it. I certainly cannot use my own photos to infer whether people’s lives generally begin to narrow after fifty. A clearer way to put it is that this organizing pass reminded me that even years I had considered uneventful still contained fragments worth revisiting.
The curve can help me ask questions, but the answers still have to come from the photos and my own memory.
The Clues and Limits of Metadata
Sometimes my memory of the season of a trip is vague, while the time recorded in a photo offers another clue. Placing several images in chronological order can also help reconnect a fragmented memory.
But metadata still has to be interpreted. Fields may be missing, time settings may be wrong, and files from different sources may not preserve information in the same way. ExifTool’s official documentation describes how date and time fields are read and written, including differences in how some formats preserve time zones. These fields can be written and changed. They are not witnesses that can never be wrong.
I therefore no longer describe metadata as more honest than memory. Its value is that it gives me another clue I can return to and check. When memory and a file disagree, I can first ask which field supplied the time, whether location data exists, and whether the image is an original or a later copy.
Seen this way, organizing photos means building an index for my own time. An index can help me find a fragment, but it cannot decide what that fragment means in my life.
From Memory to Query
Images, cloud files, social records, and health data make it possible to search and compare some clues that once could be found only through recollection. This works only if the data still exists, can be accessed, and is understood for what it actually records.
For example, I once considered comparing photo timestamps with Fitbit records. That could be one direction for comparison, but I would first have to confirm that the data was available, that the times and time zones aligned, and how long the sampling intervals were. A record close to the time a photo was taken is not necessarily the state of my body at the exact instant I pressed the shutter, and it certainly cannot explain how I felt in that moment.
When files become searchable, I gain another way to relate to the past. Alongside “I remember,” I can ask, “Is there a record I can check?” But failing to find one does not mean the experience never happened.
This reminds me of my experience rebuilding a website with AI. Organizing experiences and ideas into content that can be found reduces the difficulty of searching for them again later. Photo organization is similar, except that images often involve both my life and other people’s lives. Before giving them to a tool or sharing them, I ask what data is needed, who can see it, and whether locations, people, or health information extend beyond the purpose of the organizing task.
An Index for the Future
Ten or twenty years from now, when I want to recover a fragment, what can the present version of me leave behind?
Rather than pursuing an agent that contains all of memory, I would first preserve readable files, their sources, and the method used to organize them, while also caring for backups and access permissions. Data structures require maintenance. Being searchable does not mean they will last forever.
The number 70,267 once made me stop and look at my life. What I most want to preserve, however, are the paths that can lead me back to the photos.
Images may retain a time and place, but they do not automatically know whom I missed or what mattered to me. I still need to write or say those things, or understand them again when I return to the images later.
Having a way back to a fragment is already reason enough to organize it. I can leave the rest for my future self to fill in slowly.
Revised November 2026: This revision preserves my personal experience of organizing those photos while clarifying that image counts, metadata, and memory can complement one another, and that each also has limits.


💬 Comments
Loading...