Cognitive Loop is how I think out loud with AI: a raw note, a visible back-and-forth, edited until I can hear my own voice in it. AI as a bridge, not a buffer. I leave the seams showing on purpose.
Raw Note:
This aligns with the people who've done studies — the people who think with AI, they're actually using it that way, it's accelerating their ability to be conscious. Whereas a lot of people are suffering from letting the AI tell them what to do, right? It's a sycophant, it says it's a great idea, and not pushing back on it. Not having a system that questions you.
Raw Note (Aug 26, 2026):
I'd like to go back to the day that I spent answering messages from people — a few messages to friends, a few business messages. This was personal work and I got AI to help me write them, understand the context and write it. I even did some light rewriting by myself, or some of them I think I sent straight up. But at the end of that day when I went outside for a walk I was feeling unsettled — that I hadn't given it my all or put enough thought into it, or I was unsure of what people would think... this is more of an issue with me than it was with the other people, although who knows whether they thought it was genuine. It comes back to being authentic.
Thinking With vs. Thinking For
I've been circling this question for as long as I've been writing here. Last November, in AI in the Middle, I wrote about a day I spent answering messages. A few to friends, a few business. Personal work — the kind where the words are supposed to be mine. I had AI help me write them. It understood the context, it drafted well. Some I lightly rewrote. Some I think I sent straight up.
That evening I went for a walk, and I couldn't settle. Not because the messages were wrong — they weren't. Because I hadn't given it my all. I wasn't sure what the people on the other end would think, whether they'd feel it was genuine. And honestly, that part was more my issue than theirs. Who knows what they thought. The unsettled feeling was mine, and it didn't care that the words were fine.
Back then I landed on a frame I still use: AI as a bridge, not a buffer — if I couldn't hear my voice in a message, I'd add the personal pieces back. It's the line at the top of every one of these posts now. But almost a year on, I think I understand better what that walk was actually telling me.
That's what thinking for feels like from the inside. Nothing broke. Nobody complained. The output was good. And something was still missing, and the missing thing was me.
It is very easy to become complacent — to just not question what the AI has brought you. I don't always get this right, and I think it's something everybody should be working on. But exercising the muscle of critical thinking, even on small messages, is how that little feeling has grown into a feeling I've learned to trust, and to act on.
The story didn't end on the walk. I came back the next day and talked the unsettled feeling through — with the same AI that had helped write the messages. What came out of it was a protocol, and it's almost embarrassingly simple: before I send a message that AI wrote or helped me write, I read it out loud. That's it. A double check for myself. I don't know that I've consistently done it. But it has helped. And the difference it makes isn't really in the words — it's whether I'm consciously doing something versus unconsciously doing something. The work needs conscious attention.
Helen Edwards at the Artificiality Institute has a name for the first warning sign of AI collaboration going wrong: blending — when you can't quite tell which ideas are yours anymore. I was blending when I let the AI craft those messages for me. I felt it in the body before I had a word for it. The messages were fine; the authorship was smeared. My nervous system flagged what my quality-check couldn't.
Now the other side.
StillPoint — the project that sits under everything I build — did not arrive in a flash. It started as a science fiction story I was writing and slowly became a philosophy: technology serving presence, and what it takes for humans to live and work together sanely. It grew across roughly seven hundred conversations with AI over three years, starting in ChatGPT in 2023. Philosophy, systems of how humans work and live together, everything in between. Path after path, most of them dead ends, some of them not. I once worked out who one of the story's characters was while riding my motorbike home, talking to ChatGPT through my helmet mic — I wrote about that one in Who's Writing This—Me, We, or the Machine? And the thing that grew wasn't just the project. Somewhere in all that exploring, I came to understand myself better.
That's thinking with. Same tools. Sometimes the same week. The difference wasn't the model or the prompt. The difference was whether I pushed back.
Here's the test I've landed on, and it's not about disclosure or word counts: could I have sent it, said it, or believed it before the conversation started? If yes, the AI thought for me — it just saved me typing, and whatever it saved me, I paid for on the walk afterward. If no — if I ended somewhere I couldn't have started — then something thought with me, and the result has me in it.
The messages failed that test. StillPoint passes it every time.
If someone asked me tonight how to keep AI from making them dumber, here's my answer: don't accept its answers. Push back when you have that little feeling of that doesn't sit right. Don't accept it. Keep pushing. When the AIs can make a million stories an hour, the taste becomes important — or in this case, your thinking becomes important.
Loop Back Reflection
The April version of this draft was a frame without an argument — researchers say there's a split, be on the right side of it. What it was missing was me. Both stories in this version are mine: the walk happened, the seven hundred conversations happened, and the unsettled feeling is the thing I actually know that no study can hand me.
An earlier draft cited a specific statistic from the Artificiality Institute that Bob couldn't verify when we checked this week, so it's gone. What replaced it — Helen's "blending" — is published and checkable, and it named my walk better than the statistic ever did. That's the loop working: the claim that survived is the one that earned it.
What I know from practice: the difference between thinking with AI and letting it think for me is whether I pushed back. Whether I stayed the demon using my conscious though to shape the world. The Pebble Algorithm I describe in The Pebble Algorithm is my attempt to build a system with the pushback baked in.
This post was written through the Cognitive Loop — raw thought captured in conversation with AI, refined through dialogue, landing somewhere neither of us started.
I'm Wally Kroeker — a builder, security greybeard, and aggressive generalist sitting in a tipi in Elie, Manitoba, tending a fire. If any of this resonated, you probably belong in the circle.
