Every component in the system had a biological name — except agents.
Tools were enzymes. Skills were receptors. Cell types mapped cleanly to agents in post #14. But that created a problem: “cell type” is a multicellular concept. Macrophages and T-regulatory cells make sense inside a body with organ systems and immune coordination. A single-celled organism doesn’t have cell types. The naming test failed.
Three attempts before it worked.
Attempt 1: Cell types (macrophage, treg)
Clean on the surface. Agents specialize, operate autonomously, report through signaling channels — macrophages do exactly this. But cell types are a multicellular abstraction. The organism I’m building is one cell that produces outputs. Importing multicellular vocabulary at the wrong level breaks the constraint.
The biology said: wrong scale.
Attempt 2: Fission
Bacterial fission produces two identical daughter cells. When a CC agent is spawned, it gets its own process, runs independently, terminates. Sounds right. But agents get different prompts. They’re not identical copies. Fission implies genetic equality — both daughters carry the same DNA, run the same programs. Agents don’t.
The biology said: wrong mechanism.
Attempt 3: Biofilm (for teams)
Biofilms are colonies of genetically similar bacteria that coordinate behavior through quorum sensing. Agent teams coordinate too. But biofilm assumes genetic similarity — the bacteria share almost all their code. In an agent team, different agents run different models. The diversity is the point.
The biology said: wrong premise.
The fourth attempt: yeast
Yeast is a single-celled organism that reproduces through budding. The bud forms on the parent cell’s surface. It grows. It receives a copy of the genetic material. Then it detaches and operates as an independent cell — smaller than the parent, autonomous, running its own metabolism.
This is what Claude Code agents do.
A CC agent is spawned from the parent process. It receives instructions (its “genetic material” for that run). It executes independently. It terminates and returns its output. The parent doesn’t need to be involved during execution.
Bud. Not cell type. Not fission. Not biofilm.
The design question the name generated
“Agent” never generated this question. “Bud” did immediately: in yeast budding, the bud inherits the full genome. Does a CC agent inherit the constitution?
Currently: no. Agents are born without it. They get their task prompt, and that’s the extent of their inherited material. They don’t know the organism’s naming conventions, the design constraints, the rules that every other component runs under. They’re blank cells that know only their one job.
No wonder agents occasionally violate conventions. They weren’t given the DNA.
The fix is mechanical: wire agent spawn to automatically inject the constitution into context. Every bud carries the genome. This is not an optional enhancement — it’s the biological imperative. A yeast cell that buds without passing on the genome produces a non-viable offspring.
The meta-lesson
Three failures before the right name. The failure pattern was instructive:
- Attempts 1 and 3 used the wrong scale (multicellular when the system is unicellular)
- Attempt 2 used the wrong mechanism (identical copy when the outputs differ)
The fourth worked because yeast budding is the only common biological mechanism where a unicellular organism produces offspring that are both genetically initialized AND operationally distinct — smaller, different stage, potentially different expressed genes.
The break wasn’t in biology. It was in my search. I kept looking at multicellular systems because “agent” carries connotations of specialization and hierarchy — which are multicellular concepts. Once I constrained the search to unicellular reproduction, the answer was one organism away.
Persistence in titration pays off. But persistence alone isn’t the lesson — the failures had to be read correctly. Each one narrowed the search. Wrong scale. Wrong mechanism. Wrong premise. The fourth attempt worked because the prior three had already ruled out everything that wasn’t yeast.
The naming discipline includes knowing when you’re wrong. “No good biological name” three times in a row isn’t a signal that biology lacks the concept. It’s a signal that you’re searching at the wrong level.