The Rock That Bounces Off Water

A few months ago I walked into a customer meeting. It was a new guy I never met before. We started off by usual pleasantries about how the day was going, but then the time came to talk about business. I stumbled. What the hell is he building? Just yesterday I asked my AI agent to help me prepare for the day’s meetings - collect all it knows about the customers I am meeting and give me a brief summary. But I could barely remember which one I was talking to. Was it temporary amnesia or something else?

I made it through the meeting and then I sat back and reflected. And I realized something. I did not forget details about this customer. I never knew them in the first place. AI did the heavy-lifting for me, it searched the Internet and gave me a well-cooked answer. I borrowed the knowledge. It is like a rock that bounces off water. The summary touched the surface, nothing sank.

Cheating on an exam works the same way. You answered the questions correctly, but you learned nothing. The next day you can never answer any of the questions from the exam. Effort is like an encoding mechanism for knowledge. It transforms what you read or hear into brain signals that stick. If you don’t struggle to understand, to connect the dots, you don’t retain it.

Maybe this argument rings a bell. 2,400 years ago Socrates argued that writing will create a civilization of shallow people who really know nothing. And it echoed many times. What stops costing effort, atrophies. People used to remember dozens of phone numbers. Now can’t cite a single one and it’s ok. On the other hand, people could somehow navigate huge cities like LA by memory, truckers could drive coast to coast without needing a map. Now we open Google Maps just to go fetch groceries from a store 5 blocks away. What is the difference?

The difference is structure. Phone numbers are disposable knowledge. Same as fun facts about Venus. What is its diameter - 3,000 km or 5,000? Who cares! This is information that can easily be offloaded and stored somewhere else until we really need it. Like constants in physics. Just look it up in an encyclopedia. But understanding city or road grid is different. You can look up the highway number, but you still need to know where it goes. That’s structural knowledge. An AI summary is rented knowledge. Writing it yourself is owning it. Renting is fine — until you rent something load-bearing.

So now all my customers get the by-hand treatment. I open the customer’s website, read their mission, check their blog posts. Look up LinkedIn profiles to put a face to the name. I summarize it myself, in a format that works for me, connecting the dots. When AI made research feel easy and removed friction, it also removed learning. But my AI assistant does not sit unemployed, it does an important job for me, the one where it excels - filtering through the noise. Reading product announcements in endless Slack chats, getting up to speed with what the competition shipped - it is all handled by the machine first. Most of that I would forget by Friday anyway. And that’s the point.

The rock has to break the surface to sink; skipping across doesn’t count — no matter how many times it touches the water. Socrates was wrong about writing. Am I wrong about this?