I had twenty-six thousand emails in my inbox.
I was an early adopter of gmail. With it came decades of accumulated digital debris. The number came up the first time I connected my inbox to Claude.
The connection was the smallest move I had made yet in this whole experiment. A few clicks. A consent screen. A confirmation that the integration was live. I asked Claude to scan the inbox and label what it found. Newsletters here. Receipts here. Mail from people I actually know here. Threads that had gone stale and could be archived here. A first pass at the inbox, like asking a friend who is good at organizing to come over and help you with the spare room.
It worked. Within an hour, the inbox had categories I had never given it. Within a few days, I was using the categories to make decisions about what to read first in the morning and what to ignore. The grunt work of inbox management, the kind of work I had been doing badly for fifteen years and feeling guilty about, had been done.
That was the entry point. Inbox triage. The least interesting use of an email account I can think of.
What it taught me was that the integration could read.
The next ask was a search. I was looking for an old email, with an attachment I needed except I had no idea when I received it or from whom. I described it to Claude in plain language. Claude found it in twenty seconds. The email was from 2014. I had been looking for it on and off for months.
That was when I understood what was actually in front of me. Not an inbox tool. A search engine over my own correspondence. Plain language in, an answer out, with a pointer to the actual evidence behind it.
Later that day, I was working on a manuscript I have been writing for several years. It is a memoir. I had reached the part of the work where I was no longer drafting and was instead gathering. Pulling together every place I had ever recorded a thought about the subject, so I could see what I had and decide what was missing. Blogs. Photographs. Handwritten letters. And then, I realized, emails. Seventeen years of emails.
A separate email account that I had used, since 2009, for all correspondence related to adoption. My own adoption story as a parent. The friends and strangers who had helped me think through what I was doing. The records that had shaped my understanding of what it meant to raise children whose origins were not mine. Hundreds of messages. People I had not spoken to in years. Conversations I half remembered. Attachments I had not looked at in a decade.
I gave Claude access to that account.
We scanned it iteratively. Person by person. Thread by thread. I would ask, what do we have from this person, and what was the arc of our exchange. Claude would surface the thread, summarize what was in it, point me at the messages worth re-reading. I would read them. I would decide what mattered. I would tell Claude what to note for the timeline we were building. We would move to the next person.
Over a few sessions, what had been a cluttered inbox of stale correspondence became a structured timeline of fifteen years of one part of my life, with names attached, dates attached, the contents of attachments transcribed where they mattered, and pointers back to the original messages whenever I needed to verify a detail.
I had been trying to do this work by hand for years. I had a folder of notes. I had a partial chronology. I had the emails themselves, and no method for making them speak to each other across a decade and a half.
What changed was not that the work got easier. The work got possible.
The emails were the beginning. The attachments were the second realization.
When you scan an inbox for content, you are not just scanning text. You are scanning every file that anyone ever sent you. PDFs of court documents. Photographs of paperwork. Scans of handwritten letters that someone had photographed because they did not have a scanner. Spreadsheets. Voice memos as MP3 attachments. Images of forms with fields filled in by hand. All of it had been sitting in my inbox for years, and all of it was now part of the searchable record.
Among the various artifacts I was gathering to rebuild my memories from that many years ago, was a handwritten letter I had written to my husband on the night I decided adoption was the way forward. In it, I laid the case of why we should adopt. I knew that this was something I would want to include in my memoir but I wanted it transcribed. In 2009, I was writing cursive and cursive when you are emotional does not read too well. So it sat.
I added the letter to the pile of images to be processed. Not by me. By the system. The contents were transcribed into the timeline next to the date the letter had been sent. I read the transcription. I went back to the original letter to verify a phrase. The letter had been waiting for me for fifteen years, and the friction that had been keeping me from it was gone.
Multiply that by every attachment in fifteen years of correspondence. That is the second realization. The inbox was not the archive. The inbox was the doorway to the archive, and the archive was much larger than I had been treating it as.
The third move was outside email entirely.
By the time the inbox work was running, I had developed an instinct for what I now wanted to feed the system. Anywhere I had ever recorded a thought, sent a message, saved a document, taken a photograph of something I might want to remember. All of it became eligible. The question stopped being what data do I have and became what data can I get to.
I started with chat exports. Years of conversations that lived inside messaging platforms I had drifted away from. Some of the platforms made it easy. They had an export option. You clicked it, you waited a few hours, you downloaded a file, you handed the file to Claude. The conversations became part of the timeline.
Some of the platforms did not make it easy. The export option was buried, or partial, or deprecated. For those, I started taking screenshots. Long conversations require many screenshots. I would scroll, capture, scroll, capture. I would hand the screenshots to Claude, in batches, and ask for the contents to be transcribed and added to the timeline.
Some of the conversations were too long for screenshots. A thread that had run for years did not fit in any reasonable number of stills. For those, I started recording my screen. I would open the conversation, hit record, and scroll slowly from the top to the bottom. The recording would be ten minutes long. I would hand the video to Claude. Claude would slowly extract the text from the video and add it to the timeline.
A reader who has not seen a tool do this is going to pause at that paragraph. I want to be honest about how unusual it felt the first time I tried it. I was sitting at my desk. I had a video file of myself scrolling through a conversation. I uploaded it like I would upload a PDF. I described what I wanted. The system read the video. It returned a transcription. I checked the transcription against my memory of the conversation. The transcription was right.
That was the moment I understood what data-format-agnostic actually meant. Not that the system could read several formats. That the system did not care about the format at all. If I could capture it as a file, it could read it. The format of the evidence had stopped mattering.
What this meant for my research, in practical terms, was this. The constraint was no longer can I get this into a form the tool can use. The constraint was can I find it at all. Anything I could put my hands on, anything I could record, anything I could photograph or scan or export or screenshot. All of it was research material now. The doorways had been propped open in every direction.
I want to be careful here. The tool reading my correspondence is not the same as the tool understanding my life. I read the transcriptions. I made the decisions about what mattered. I built the timeline by hand, choosing what to keep and what to leave out, what to cross-reference and what to let stand alone. The tool was the librarian. I was the historian. The distinction matters and I am holding it firmly.
You handed a tool access to your inbox. To a second inbox holding fifteen years of correspondence on the most personal subject in your life. To attachments. To chat exports. To screen recordings of conversations you had been part of. The amount of access you describe is staggering. How could you have done that?
The honest answer is that I have been doing some version of this for twenty-five years.
The notes for everything I have ever written are on a desktop computer. Sensitive correspondence has lived in cloud-based email since the early 2000s. Photographs of my children have been backed up to servers I do not own. Manuscripts have been drafted in software that syncs to the internet by default. Tax returns have been filed through online portals. Legal documents have been signed with services I clicked through without reading the terms. The intimate documentation of one writer’s adult life has been distributed across the storage infrastructure of half a dozen large companies for as long as I have been an adult on the internet.
The category of trust the post describes is not new. The scale is. What is staggering is not that a tool can read my correspondence. It is that a single tool can read across all of it at once and make connections I could not have made by hand. That capacity is what feels different. The underlying question of whether to trust a digital system with sensitive material was settled, by me and by most adults reading this, a long time ago.
I also chose the technical posture deliberately. The work I have described in this post happens largely inside Cowork, an environment that is sandboxed and where the files I am feeding to the system reside locally on my own machines. The sandbox does not eliminate trust. It bounds it. The tool sees what I show it. It does not have ambient access to my computer, my browser, my other accounts. When I close the session, the session ends.
The question is not whether to trust digital tools with sensitive material. The question is which tools, at what depth, with what safeguards, in service of what work. Those are the questions I have been answering, by hand, for the last several weeks, and I expect to keep answering them for as long as I am a writer with a digital archive that holds the material my work depends on.
The AI Journey
- The Morning I Downloaded VS Code — The morning I opened the terminal I had walked away from twice.
- Two Hours on a Monday Morning — A publishing pipeline, five weeks of failure, and the fifteen-minute replacement.
- Nine Point Six Percent — The SEO audit that asked me what twenty years of writing were for.
- The Librarian and the Historian — Reading across fifteen years of correspondence for a memoir.
- What the Dashboard Doesn’t Count — The cost of the migration that went wrong.
- Saturday Morning — The practice I have concluded for myself.
Leave a Reply