Programming is becoming a form of exploratory conversation.
Software increasingly begins as conversation. Source code is no longer always the first formal representation of an idea; it is often the compiled form of an extended dialogue.
That dialogue may begin as a loose description, an analogy, a question, or a half-formed idea spoken aloud. Through conversation, the idea becomes clearer. It develops boundaries, behavior, relationships, and purpose. Only then does it harden into architecture, interfaces, tests, and code.
intuition ⇄ conversation ⇄ clarified intent ⇄ system design ⇄ software
This is not code-free development. Code still matters, and reality still gets the final vote. It is better understood as code-indirect development: human intention moving through natural language, models, tools, formal languages, and runtime behavior.
The conversation is the first draft
Conversational development allows an idea to remain fluid longer. Discovery happens while the idea is being expressed. The developer is not merely describing a solution; the solution is being found through description, critique, testing, disagreement, and revision.
The quality of the finished software increasingly reflects the quality of the complete conversation—not the cleverness of one prompt.
Curiosity gets something to push against
The small games began as experiments with the frontier models available at the time. Their code was entirely generated from prompts and subsequent guidance. My involvement was asking, trying, testing, and reporting back. I wanted to see what those models could make and what the results would teach me.
That made an idea easier to encounter. I could press a button, move through a little world, and discover where my description was incomplete. Later revisions belong to that record too; an artifact can preserve an earlier capability while acquiring new behavior.
I Was the Crash Test Dummy records that distinction. The September excavation follows the same curiosity into shared meanings, aging evidence, and operational memory.
An ecology, not an oracle
These experiments are rarely the product of one model or tool. ChatGPT, Claude, Gemini, NotebookLM, Grok, VS Code agents, image generators, music generators, tests, users, and runtime evidence contribute different strengths and different failure modes.
The useful question is no longer Which model is best?
Which arrangement of people, tools, perspectives, and evidence gives an idea the best chance to become reliable software?
No model receives unlimited trust. The human retains responsibility for purpose, boundaries, coherence, and acceptance. Models propose and challenge. Tests verify claims. Runtime reveals what is actually true.
An emerging practice
I am new to cybernetics as a formal tradition, but not to the patterns it describes. I arrived through practice: building systems with feedback, watching authority move between people and tools, preserving disagreement, and learning that trustworthy systems must be able to notice and correct their own errors.
This is not a finished doctrine. It is an R&D notebook for an immersive learner documenting a developing practice of conversational cybernetics through distributed judgment.
More room for curiosity
AI does not create more hours in the day. It changes what those hours can contain.