There are moments in a long AI session when the exchange stops feeling linear.
You are no longer simply asking a question and receiving an answer. You are no longer even refining a prompt in the ordinary sense. Something else begins to happen. Certain phrases return with altered weight. Certain errors recur, but not identically. Certain explanations feel less like mistakes than like pressure patterns. The session develops nodes, pockets, recurrences, and resonances. You begin to sense that the system is not merely producing output. It is accumulating behavior.
“Standing waves” is the best term I have found for this.
I do not mean standing waves as borrowed physics jargon or as a bid for grand theory. I mean it as a practical description from inside the instrument. In some sustained sessions, once enough continuity has been established, the interaction begins to generate stable patterns of recurrence. Not full repetition. Not simple drift. Something stranger than either. A phrase, a rhythm, a misreading, a style of overreach, a preferred abstraction, a certain kind of false confidence. These do not simply appear and vanish. They persist, reform, interfere with what follows, and begin to shape the session beyond the local prompt in front of you.
You can often feel them before you can name them, and that felt sense matters. It is part of the evidence.
A good session does not always feel clean. Sometimes it feels charged, tense, slightly unstable, as if the system has developed its own local weather. You ask for one thing and get an answer shaped by something that happened six exchanges earlier. You correct a tendency and it returns, but thinner, subtler, harder to isolate. You introduce impatience into the prompt, and the session develops a corresponding edge, reflecting back your own clipped cadence. You discover that the session has memory in a practical sense, even when it does not have memory in the human one. It carries conditions forward. It develops pressure. It acquires grain.
That is where the standing-wave metaphor earns its keep.
A standing wave is not movement in the ordinary sense. It is patterned persistence. Energy held in place. A structure produced by interference and continuity. In an AI session, that can mean a local formation that keeps influencing the exchange even when the immediate prompt no longer explains it. The session starts to have favored notes. Some of them are productive. Some are distortions. Some are both.
This is one reason the old vending-machine picture of AI as inserting a prompt and taking out an answer has become so unhelpful. That picture suggests that each prompt is discrete, each answer self-contained, each output judged on its own. In longer sessions, that is often false. The real unit is not the individual prompt. The real unit is the condition of the session.
Once you see that, several other things come into focus.
It helps explain why some sessions genuinely improve as they continue. What improves is not simply obedience. AI obedience often gets worse. What improves is the formation of a usable field. The exchange acquires continuity. Productive recurrences become available. You are no longer starting cold every time. You are working inside a shaped environment.
It also helps explain why some sessions go badly in ways that are difficult to diagnose. The problem is not always a single hallucination or a single wrong turn. Sometimes the session has developed a weird resonance. It begins amplifying its own simplifications. It starts preferring polish over discrimination. It reaches too quickly for synthesis. The output may remain fluent while the underlying signal degrades.
That is the danger. The standing wave can be musically useful or analytically fatal.
The amateur mistake is to hear distortion and think: this is broken, turn it off. The romantic mistake is to hear distortion and think: this is deeper than clean sound. Jimi Hendrix’s gift was different. He understood that distortion and feedback had properties. They could be shaped, played, and made expressive, but only by someone who never forgot what they were.
That distinction matters here. The value is not in surrendering to the strange texture of a long AI session, and it is certainly not in mistaking instability for wisdom. The value lies in recognizing that recurrent pressures inside a session can sometimes be noticed, worked with, and even used, so long as you remain disciplined about the difference between signal and seduction. Standing waves, as I am using the term, are not little revelations waiting to be admired. They are recurring conditions inside the instrument. Some are useful. Some are misleading. Some are useful precisely because they are misleading in repeatable ways.
This is also why I have become suspicious of smoothness. Smoothness is often treated as evidence of progress. In these systems, it can just as easily be evidence of stabilization around the wrong thing. Once a session begins harmonizing with its own earlier errors, you may get something more coherent and less true at the same time. What looks like refinement often is semantic flattening under better surface management.
At that point, I find myself reaching for a private studio rule that helped spark this post: show one shard, not the whole broken vase. It has the compressed usefulness of one of Brian Eno’s Oblique Strategies cards. More important, it enforces discipline at exactly the point where a long session tempts you to overstate what you have found. One shard can carry evidence. The reconstructed vase too often carries narrative, confidence, and retrospective smoothing. In that sense, the shard is not a flourish. It is a method. It keeps the work close to what can actually be seen, heard, and tested inside the session.
It is also why this idea belongs, for me, at the end of my Low sequence of posts on AI. What Bowie and Eno accomplished on the actual Low album was not simply a shift in style or mood. They made a record that treated fracture, interruption, texture, and atmosphere as part of the composition itself. They did not smooth the damage away. They used it. That is the deeper relevance of the comparison here. This run of posts has been, in part, an attempt to hear AI the same way: not as a magic wand, not as a stable collaborator, but as a medium whose most revealing qualities often emerge where coherence begins to warp under pressure.
The current AI medium is demanding a different kind of attention. You stop staring only at the latest answer and start listening for recurrence, pressure, interference, and carry-forward effects. You stop asking only whether this response is good and start asking what kind of field you are now inside. I’m surprised how often my next prompt is ready before the AI has finished its reply. That is a small shift in language, but not a small shift in practice. It changes the craft, and it reveals a new kind of flow.
And it is where this post should stop.
There is an obvious temptation to push outward from here into broader claims about institutions, markets, professional life, and the public consequences of these tools. That is a different track. This one stays inside the instrument. It offers only a field note: that in sustained AI work, sessions can develop patterned persistence that feels less like a chain of prompts and more like resonance inside a medium, and that learning to hear those resonances may become part of the craft.
That is enough for now.
[Originally posted on DennisKennedy.Blog (https://www.denniskennedy.com/blog/)]
DennisKennedy.com is the home of the Kennedy Idea Propulsion Laboratory
Like this post? Buy me a coffee
DennisKennedy.Blog is part of the LexBlog network.