Once Again, I Can't Find My Stuff

The Inventory I Can't Take

I tried to take inventory this week. Not of documents or files. Of AI tools.

I use Claude for this newsletter and for deeper narrative work. I use ChatGPT for other projects. Gemini shows up in Google whether I ask for it or not. At work, there's Glean, which runs on multiple foundation models under the hood. And increasingly, AI is simply embedded in products I was already using.

That's six or seven AI relationships running simultaneously, and I'm probably undercounting.

And then there are the tools I chose years ago—Evernote, Dropbox—which have added their own AI features recently. So make that eight or nine.

What struck me wasn't the number. It was that each one knows something about me that the others don't. 

Claude has context about how I write, how I think about narrative strategy, what I've worked on for over a year. ChatGPT has a different version of me—different projects, different questions, different patterns and habits. Gemini knows what I search for. Glean sees how I operate inside a company–the questions I ask, the information I look for, the problems I'm trying to solve. Each tool has a different version of me.

None of them has the whole picture. And none of them talks to the others.

I couldn't tell you where all my stuff is. Actually, that’s not quite right. I couldn't tell you where all the versions of me are.

I've Seen This Movie Before

Years ago, before this newsletter, I wrote about a different kind of sprawl. Cloud software was proliferating faster than organizations could standardize. Teams adopted whatever worked—Slack here, Trello there, Google Docs for this group, Dropbox for another. Companies scrambled to catch up—data consistency, silos, governance, security, compliance. The usual list, but at a scale nobody had planned for.

What I kept coming back to then was that the problem wasn't the tools. It was people, processes, and culture. You couldn't solve proliferation simply by choosing the right platform. You had to change how people worked.

Eventually, that wave stabilized. Slack won messaging, Salesforce won CRM, Google and Microsoft won productivity. The chaos wasn't permanent. The market consolidated, companies standardized, and the sprawl became manageable.

This feels similar. But it also feels different this time.

What's Different This Time

The last wave of sprawl was about data and operations. Every team had its own tools and workflows, its own way of storing and sharing information. The mess was real—silos, duplication, no clear system of record—but it was a mess you could see. The data existed somewhere. You could find it and move it. The problem had a shape.

What I'm dealing with now is different. It's context sprawl.

Every AI tool I use has been building a working model of me. Not a file or a document—something harder to name. 

How I ask questions, what I tend to need, the patterns in how I work, what kinds of answers I find useful. The more I use it, the better it gets. That's the whole point. Context accumulates, and the next interaction is more useful than the last one.

That problem is that the context isn't portable. I can export conversation logs. I can save transcripts. But the transcripts aren't the asset.

The asset is the accumulated understanding.

What Claude has learned about my newsletter voice, or the patterns ChatGPT has picked up from other projects. The patterns each system has inferred after hundreds of interactions.

That's locked inside each system. I could try to recreate it manually—write up my preferences, provide examples, build elaborate prompts. 

But anyone who uses these tools regularly knows that's not the same thing. 

Hundreds of interactions don't compress into a few paragraphs.

It reminds me of social media algorithms. TikTok learns what entertains you. LinkedIn learns what engages you professionally. Nobody expects to export their recommendation algorithm from one platform and import it into another.

But AI feels different. These platforms aren't learning what I watch. They're learning how I think. And the more useful that makes them, the harder they are to leave.

The danger of defaults, as I explored in Don't Accept the Defaults, isn't that they're wrong—it's that they're invisible. The default here is fragmentation. I didn't decide to distribute my thinking across five platforms. It just happened, one tool at a time, each one useful enough to justify the next interaction. And now the pieces are scattered, and nobody's offering to help me put them back together.

The Organizational Version

This is how it plays out for companies too.

Engineering uses one model. Marketing uses another. Leadership has a favorite third option. Everyone debates which foundation model should become the standard.

That's probably the wrong question.

The more interesting question is where the organization's thinking actually lives.

People are increasingly doing their best work inside AI conversations. They brainstorm. Analyze. Explore alternatives. Work through ambiguity. Pressure-test ideas. The conclusion might show up in a deck, a document, or an email.

The reasoning often doesn't. The reasoning stays inside a chat thread attached to an individual account.  That's a different kind of knowledge-management problem.

I once described Narrative Debt as a house where every leader adds a room without checking whether it connects to the existing structure. This is Narrative Debt made literal.. Same thing, except now the rooms are invisible. 

The rooms aren't disconnected.

They're invisible.

When someone leaves, the organization keeps the outputs. But much of the thinking that produced those outputs disappears with them.

Awareness is the first casualty of convenience was about how automation replaces noticing—small decisions you used to make consciously get absorbed into systems until you forget they were decisions at all. That's what's happening here. Each time you use an AI tool, you're making a decision about where a piece of your working intelligence lives. You're just not thinking of it that way.

What I Don't Have an Answer For

The last wave of software proliferation ended with consolidation. Categories settled, standards emerged, and companies picked a stack and committed. Maybe this one will too.

Some AI tools will become features inside larger platforms. Others will emerge as foundational layers. Standards will develop. Markets usually sort themselves out eventually.

I don't know if this wave ends the same way. Some of these tools will get absorbed by incumbents—the way so many productivity apps eventually become a feature in some established enterprise or consumer software. Others are competing to be foundational. The shakeout hasn't happened yet.

What I keep coming back to is that the switching cost this time isn't data. It's the way a tool has learned to think with you—the accumulated context that makes it useful in ways a fresh install never could be. 

The thing that makes a tool valuable is the same thing that makes it difficult to leave. The fluency and the lock-in are the same thing.

People are working on portability—tools that transfer chat history, platforms that manage enterprise context, systems that have shared context. But moving transcripts between tools isn't the same as moving understanding. That part still feels unsolved.

Berkson's Bits

When you're talking to an advisor, pay attention to the questions THEY ask YOU. Those questions ARE often the advice."

What I’m Listening To…

There is something magical about an unplugged performance by a pop star. Not all of them can perform live, but for the ones who can, it’s magical. Here’s a throwback of Tori Kelly and her guitar singing “Funny.” Beautiful and raw. Enjoy! 

I can't find my stuff. But increasingly, the stuff that matters most isn't something I can point to.

Looking forward to continuing the conversation...

Alan