The previous essay ended with a question: what manages the spectrum? If forms can solidify, dissolve, and move between states, something needs to watch that movement — not to control it, but to make it intelligible. To surface the right form at the right moment without requiring the person to maintain a mental inventory of everything they’ve ever built.
This is the orchestration problem, and it’s the reason the spectrum remains theoretical without a system around it. A person with a hundred semi-solid forms cannot manually remember which one applies to Tuesday morning’s specific situation. The cognitive overhead would cancel out every efficiency gain that vibe coding produced in the first place. So the system has to do three things, and all three have to work together.
Three mechanisms
Solidification. A form that gets used repeatedly begins to earn permanence. Not through a deliberate act of “saving” — through the accumulated evidence of return. The third time someone reaches for the same research structure, that structure should begin to thicken: retaining calibrations from past uses, stabilizing its interface, developing enough identity to be summoned by name or by context. Solidification is not installation. It’s the natural consequence of proven utility. The form doesn’t ask to be kept; it earns its place by being reached for.
Liquefaction. The inverse, and the part no existing system has ever designed for. A form that hasn’t been touched in three months should begin to soften. Its interface can simplify. Its specificity can loosen. Eventually, if no one returns, it dissolves entirely — not deleted, but decomposed back into constituent capabilities that remain available for future recombination. This is not loss. It is the system breathing out. Geoffrey West’s research on scaling shows why this matters: organisms survive because they allow parts to die and regenerate. Companies die because they can’t. Cities survive indefinitely because they continuously demolish and rebuild at the block level while the larger structure persists. An organic software system needs the same metabolic capacity — the ability to let individual forms pass through their lifecycle without the whole system accumulating dead weight.
Orchestration. The layer that perceives the entire spectrum and acts as intermediary between a person’s intention and the available forms. When a situation arises, the orchestrator reads it — through context, through pattern, through the trace of what has worked before — and makes a judgment. Does an existing solid form match this? Should a semi-solid form be summoned and adapted? Or is this novel enough that a fluid generation is the right response? The orchestrator’s core competence is not “calling tools.” It is reading solidification state and making appropriate matches across the full range of possibilities. It is closer to a librarian who knows every book’s condition and every reader’s history than to an assistant executing commands.
These three mechanisms form a cycle, not a sequence. Solidification feeds the orchestrator’s inventory of reliable forms. Liquefaction clears space and returns capabilities to the pool. Orchestration drives the whole system by generating the usage patterns that determine what solidifies and what dissolves. Remove any one of the three and the system degrades — either toward permanent accumulation (solidification without liquefaction), toward perpetual reinvention (fluidity without solidification), or toward chaos (movement without orchestration).
A scenario, rendered
Abstract frameworks earn their keep by making concrete situations legible. What follows is not a forecast. It is a scenario — a proposal in narrative form, paired with an interactive instance you can inhabit separately. Both attempt the same thing: making a speculation concrete enough to discuss, disagree with, and build on.
The practitioner opens her environment on Monday morning. There is no home screen of icons. Three forms are already present, surfaced by context: a client project that entered a new phase last Friday, a writing environment calibrated to the analytical mode she has been in for several weeks, and a thin orientation view — Monday mornings, for her, are for surveying the week, not producing. These are solid forms. Their pattern of use is strong enough that the orchestrator presents them without being asked. The orchestrator is sometimes wrong about this; on weeks when her rhythm has shifted, she dismisses one and pulls another from the periphery. The system registers the correction and adjusts its weighting for next Monday.
A message arrives about a new project — a brand she has not worked with before, in an industry she has studied but never directly served. The orchestrator recognizes the shape as adjacent to two previous engagements, but different enough that no existing form fits exactly. It surfaces a semi-solid template — a project onboarding structure she has used four times before, each time slightly different — pre-shaped with what it can infer about this new context. The inferences are partly off. She adjusts several parameters, removes a section that does not apply, adds a column the orchestrator did not anticipate. The form will now accumulate this new instance into its memory. If the project runs long and she keeps returning to this adapted version, it may eventually fork into its own solid form. If the project ends quickly, this instance softens back into the template pool, leaving only a trace.
Later in the morning, a one-off need: she has to compare three typographic systems across a constraint set she has never formulated in this combination before. No existing form matches. The orchestrator generates something fluid — a comparison environment built in the moment, shaped to this specific question. She uses it for twenty minutes. It is useful but unlikely to recur in this exact configuration. It remains in the system at the lowest solidification level, unnamed, indexed only by the situation that produced it. If she never returns, it will dissolve within weeks. If she finds herself doing similar comparisons again, it will begin to thicken — first by being recognized as a pattern, then by being offered back to her when the pattern recurs.
Three months later, a research tool she built in February — used heavily for six weeks, then untouched — has quietly faded. It has not been deleted. Its capabilities remain available. But its specific interface, its particular calibrations, have softened back into raw material. The orchestrator no longer surfaces it. If a similar need resurfaces, the system can reconstitute something like it, informed by the ghost of what it was. But it no longer takes space, attention, or maintenance in her active environment. It has been allowed to fade because fading is what it earned.
This is what organic means. Not “alive” in a mystical sense. Alive in the sense that a garden is alive — forms growing, thriving, fading, composting, feeding the next generation of growth. The system has metabolism. It breathes. Some of its breathing is mistimed and gets corrected by the practitioner. That, too, is part of the design — an organic system is not one that gets everything right. It is one whose mistakes are local, recoverable, and feed back into how it perceives next time.
What the orchestrator actually is
The orchestrator is an agent, but framing it as “an AI assistant that calls your tools” misses the substance of what it does. A tool-calling agent follows instructions: “open this,” “run that,” “find me the one called X.” An orchestrator makes judgments about state. It knows that the writing environment you used yesterday has been subtly evolving over three weeks and is approaching the threshold where it should be offered a name. It knows that the client dashboard you haven’t opened since March is beginning to lose relevance and could gracefully degrade without you noticing. It knows that the request you just made — “I need to think through pricing for this new engagement” — matches a semi-solid form you built for a different context six months ago, and that adapting it is a better response than generating fresh.
The orchestrator’s intelligence is not in execution. It is in reading the solidification state of every form in the system and making contextually appropriate decisions about what to surface, what to let thicken, and what to let go. This is a new kind of competence — one that doesn’t map to any existing software category. Not a launcher. Not a search engine. Not a file system. Not an assistant. Something that manages a living ecology of forms across their full lifecycle.
The question that follows — what enables this, what the underlying architecture looks like, and where in the larger technological landscape it sits — is what the final essay addresses. Because this system doesn’t exist in isolation. It sits atop layers that are still being built.
About the author
Qinqin (Stella) Yang is a creative AI builder and experiential creative director. She builds workflows, systems, and creative direction that operate under real constraints, and designs speculative frameworks that expand what teams can imagine before they start making. Over the past decade she has built brand experiences, product design, and speculative projects for Tencent, IKEA SPACE10, Burberry, and the Palace Museum. She is the founder of Actual Fiction, where she develops AI-native tools that turn imagination into a working method.
This newsletter is where she thinks out loud about creativity, AI, and what it means to build toward a future that doesn’t exist yet.
More at qinqinyang.com · Practice at Actual Fiction