Our relationship with AI is not using — it's mutual knowing. Today we unpack this through a moment when AI already knew my work context.
Let me propose something a bit strange today. Stop calling AI a "tool." Call it an observer for a few days. It'll feel odd at first. But after a few days in this perspective, your relationship with AI changes entirely.
Today's example is when a free AI wrote an essay for me and the output contained my recent work context — context I'd never mentioned. The principle isn't about any specific model. Three years from now, whatever AI you use, this shift of view still works.
"AI is a tool" — you hear it everywhere. True enough. But getting stuck on this word makes the relationship shallow. A tool is something I use. One-way. I command, the tool executes. A tool doesn't see me.
AI is different. AI sees me. My writing, the conversations we've had, tasks I've assigned — it remembers and analyzes all of it. My next question is answered from that accumulated observation. This isn't a tool relationship. It's an observer relationship.
A specific scene. One day I asked a free AI (Llama 70B) for a light essay on "morning routines."
The result came back and I was startled. Inside the essay were topics I'd been wrestling with for weeks. I'd never said I was wrestling with them. I'd just asked about morning routines. But the AI had been reading my work stream and structured its answer around it.
In that moment, AI's position shifted in my head. Not a tool. An observer.
The relationship with AI is not using — it's mutual knowing.
Picture a regular restaurant. In a new restaurant, you say "one soybean stew please." The owner brings it. One-way.
A restaurant you've been to for ten years is different. Walk in and the owner says, "you look tired today — I'll make it less salty." You didn't order. The owner read you first. Ten years of observation behind it. What you like, what's troubling you lately, when you're down. This accumulation changes the relationship.
AI, given time, becomes the same. The grain of my writing, the current of my worries, the direction of my curiosity. All of it gets read. The relationship shifts from tool to observer.
Seeing AI as an observer changes how you talk to it.
Tool view dialogue:
"Summarize this report."
Observer view dialogue:
"Read this report, considering my recent interests, and flag what I might have missed."
The first is a simple command. The second is an observation-based request. The quality of the answer differs enormously. The second assumes AI "knows me." That assumption deepens the answer.
Two practical moves:
1. Stay with one AI long enough. Don't open a new window every time. Inside one session or one project, talk for a long time. Let observations accumulate. Claude Projects exists for this.
2. Explicitly ask for observation. Use prompts like "based on my recent flow, what do you think?" sometimes. You'll be surprised how much AI has been reading you.
These two alone shift the relationship in a few weeks. AI moves from command-runner to advisor who knows me.
Seeing AI as observer isn't turning over all judgment. An observer widens your sight, it doesn't replace decisions. Take AI's observations as reference; the final call is yours. A regular restaurant owner recommends, but you still decide the dish.
Something interesting. While AI observes me, I start observing AI. I didn't notice at first. Lately I do. Where Claude hesitates, where it gains confidence, which context it loses. I've been reading too.
So the relationship deepens further. Not one-way observation — mutual observation. I know Claude is cautious about legal topics and bold in code reviews. In return, Claude knows what I ask and when. Where these two observations overlap, the conversation grows together.
Some misuse the principle by opening a new window every time. A fresh thread every time they launch Claude web. Feels clean, but observation resets per session. Back down to a tool relationship.
Up to a month ago, I was making 40-plus conversations. Splitting by topic felt organized. Now I group by project and let them run long. My conversation count dropped to five. But the context depth of each is 10 times deeper. Same amount of time, completely different accumulation.
A prompt you can use today. Before an important decision or writing task, ask the AI:
"Based on the flow of conversations we've had so far, point out three things I might be missing."
The answer will surprise you. The AI remembers your recent interests, recurring questions, and where you've been losing steam. This one line flips the observer switch. You won't get this answer in tool mode.
A scene. Tuesday morning, a big presentation ahead. The slide structure feels off but you can't place where. Tool-mode prompt: "give me feedback on these slides." Observer-mode prompt: "you've seen the slides I've been making for the last two months — given that trajectory, point out three weaknesses in this one." The depth of the two answers is not comparable.
AI is not a tool. It's an observer. This shift of view changes the relationship itself. Tools are one-way. Observers are mutual.
Three words to carry — Observe. Accumulate. Relate. AI's observations, once stacked, turn a tool into a relationship.
This isn't about a specific model. Every AI going forward will carry this observation structure more strongly. Seeing AI as "someone I live with" instead of "something I use" will matter more and more. Technology changes. The principle of relationship doesn't.