Walk First
The Setup
A few weeks ago I finished an AI-assisted deep dive across a client’s operational landscape. Twenty-eight documents. Nine markdown checkpoint versions digesting it all in batches. A hundred and thirty findings organized across eleven sections, with twenty-one open verification items to follow up with the client. It was thorough. It was structured. And when I presented it to the teams who’d actually be doing the work… deer in headlights. Nobody pushed back. Nobody asked questions. They just sat there, trying to figure out where to start. Or at least that’s how I read the room. It was mostly “wow, this is great…” and then silence. I picked up on that after… two meetings. I’d handed them an accurate map of the entire territory. The problem wasn’t the map. They hadn’t walked any of the ground yet.
My Approach + AI Role
Here’s what happened under the hood. AI canvassed twenty-eight client documents, extracted a hundred and thirty findings, organized them into a category index, and helped me build a SharePoint information architecture blueprint that went through six-plus versions as new clarity was getting unlocked between meetings and email exchanges with clients. The volume comprehension was, structurally, excellent. AI made it almost effortless to surface everything. And that’s exactly the problem I didn’t see coming. Pre-AI, canvassing twenty-eight documents would have cost teams days if not weeks. That cost was a natural forcing function. I physically couldn’t read everything fine point front to back, so I had to make choices about what mattered most for the moment. The constraint did the editorial work for me. AI removed the constraint. And I kept building. Here we go again.
What Actually Happened
Two instances. The first one I described above… the team staring at a hundred and thirty findings, not confused because the content was bad, but confused because their working memory was full before they finished the table of contents. John Sweller’s cognitive load theory says working memory holds roughly four chunks at a time. I’d handed them somewhere around thirty times that capacity and expected them to orient. The second moment happened a few weeks later on a different project. Same pattern forming. I was deep into research on a platform transition… parent company context, client pressures, positioning shifts in the AI landscape. The AI-assisted research was flowing. I could feel another interactive HTML (aka deck) taking shape. And then I caught myself. Builder Mindset going unchecked. I DID IT AGAIN. I stopped building. No HTML. No findings deck. No orientation guide. Instead, I decided to find the right moment to start the right conversation with the right person. The research stays in my head. It informs my thinking, not my deliverable. The first instance, I learned the lesson. The second time, I almost forgot it. Me catching myself every other day… that’s the whole game right now.
The Real Insight
AI doesn’t speed up the part that matters most. People still need to experience the work, make meaning from it, and build context before any of the organized information actually lands. What AI does make nearly effortless is the volume problem. And that’s what makes it tricky. The old discipline was gathering enough. You had limited time, so you triaged ruthlessly just to have something useful by the deadline. The constraint was on the input side, and it naturally compressed your output. The new discipline is refusing to deliver everything you gathered. The constraint has moved to the output side, and most of us haven’t built that muscle yet. Or I’m just a digital hoarder. Here’s what I’ve realized: structured knowledge is not absorbed knowledge. A chatbot on top of un-absorbed knowledge just creates a faster way for people to get answers to questions they don’t yet know they should be asking. I think that starts looking more like dependency than real adoption. Knowledge has three states:
- Reference (you can look it up).
- Fluency (you can operate without looking it up).
- Judgment (you can handle what the guide can’t anticipate). AI tooling solves Reference state beautifully. But if you add retrieval-superpowers on top of a team that hasn’t reached Fluency, you skip the part where they develop the Judgment to know what to ask. The fix isn’t a better interface on top of the answers. It’s giving the team the experiences that generate the questions… through doing, through time.
Try This If…
You’ve just handed someone a comprehensive, well-organized document and watched them not know where to start. Or you’re building an AI-assisted knowledge system for a team that hasn’t operated in the new environment yet. Before you build another layer on top of the information, try asking one question: does this team know what their job is this week? Not where the documents live or what it says. What they’re actually supposed to do next. Can they explain it simply back to you? If they can’t, the problem isn’t access to information. It’s orientation. And orientation doesn’t come from more content. It comes from a conversation that names what someone is already experiencing and connects it to the next step they take.
Systems Lens
We don’t lack AI tools or tutorials promising exponential gains. What we lack is the bridge from where people are now to the experience they need to make meaning from what AI can surface. I look at it this way: AI shines a light across the whole dark landscape. That’s its superpower. But just because we see the landscape isn’t the same as getting to our destination. We navigate through it by knowing where we are and seeing the next step clearly enough to take it. If AI makes the volume problem nearly free to solve, the new discipline isn’t gathering more. It’s deciding what one simple point to make so we all take that first step together.