Alberto Denia published "Code is clay now," describing a metaphor: the cost of changing code has collapsed, so code behaves like a different material even though it is made of the same stuff.
He uses two descriptions of the field: artisan and engineer.
An artisan stays in the loop and applies judgment to every iteration of the code, where an AI might be involved but the human affects every loop.
Diagram source
flowchart LR
subgraph artisan["Artisan: tight loop"]
direction LR
I1[Iterate] --> E1{Evaluate}
E1 -->|not yet| I1
E1 -->|good| S1([Ship])
end
In the other description, for the engineer, a factory removes the human from the loop: specs, generators, and verifiers generate code, arranged so that code-clay comes out as bricks that fit a larger structure, escalating to a person only as a last resort.
Diagram source
flowchart LR
subgraph engineer["Engineer: loose loop"]
direction LR
SP[Specify] --> F
subgraph F["Factory"]
direction LR
G[Generate] --> V{Verify}
V -->|fail| G
end
V -->|pass| C{Confirm}
C -->|reject| SP
C -->|accept| S2([Ship])
end
Judgment exists in both. The artisan spends it per turn; the factory spends it upstream, in the molds and the verifiers, and tries to spend as little as possible during the run.
That's not a bad set of metaphors! And he separates engineering from "vibe coding": judging only the surface is vibe coding, and judging the shape of the code as well is engineering, however fast the loop runs, or where. The kiln and the armature in his telling are taste, context, guardrails, and architecture, and clay without those things falls into shapelessness.
The factory is a good illustration of emergent structures, without the name. That piece argued that emergent design across boundaries rests on a substrate someone engineered small: narrow, legible from outside, and permitted to a second caller. The substrate is the only engineered part. Everything built on it is bricolage, a person looking at what is reachable and noticing a fit. Denia's factory is that picture with the bricoleur replaced by an agent. The molds and verifiers are the substrate. The bricks are the narrow legible surfaces. The larger structure is what gets assembled out of them to fit whatever shape the problem has this week.
The mapping also says where the factory can go wrong, which Denia leaves open. His bricks fit a structure somebody already drew, and molds cut for one building produce a substrate locked to its first caller.
That is one of the typical requirements for successful emergent design failing: a factory whose bricks only fit this building is a template engine, because it fails to permit unspecified usages.
A factory whose bricks a second caller can pick up is a substrate, and then the composition that used to happen at the speed of a person noticing something happens at the speed of an agent. The rule from the earlier piece transfers unchanged: do not design the molds for callers you do not have. Build them narrow, make them legible, and do not lock them to the first use.
It's interesting how emergent even the discussions about emergent design are becoming - Jung described this as synchronicity, and while I'm not sure Jung's collective unconscious is a mechanical manifestation, it sure feels like it sometimes.
Comments (0)
No comments yet.
Log in to comment