Morphology in AI components
An agent moves through many states as it works — thinking, streaming, handing off, waiting. This is a study of interface morphology: components that change shape to show what the agent is actually doing, not just whether it is busy.
Committed speech is solid; the trailing edge is still forming.
Here is the summary you asked for, drawn from the three most recent reports before I move on to the open questions…
Reasoning in the open, not a black box.
Checking the three linked reports for conflicting dates…
Where the human hands intent to the agent.
Summarize this thread and draft a reply
The link between two agents dropped — waiting to reconnect.
Connection lost — reconnecting to Executor…
The legacy signal: busy, but silent about state.
answer / no answer
An agent can pass through many states during a single stretch of processing — any of the states above, often several at once. Our current design language struggles to say which.
Legacy paradigms — the loader, the progress bar — encode a discrete outcome: the answer is here, or it is not. What they cannot show is the model’s confidence, what state it is in, whether it is still thinking, or whether the text on screen is partial or committed.
This experiment treats morphology — components that change shape with the agent’s state — as a way out of that binary, toward interfaces that let a person read an agent the way they read another person mid-thought.