Core Nucleus
Code that is too architecturally sensitive or security-critical for agent execution, written exclusively by human engineers.
Definition
The Core Nucleus refers to code that is too important or too architecturally sensitive for agent execution. These are the components where risk tolerance is zero: security-critical paths, core algorithms, and foundational abstractions that everything else in the system depends on.
Agent Operators write Core Nucleus code by hand rather than delegating it to agents. This boundary exists because:
- Security Sensitivity — authentication flows, encryption implementations, and access control logic require human judgment about threat models and edge cases that agents may not fully reason about.
- Architectural Foundation — base abstractions, core data models, and framework integration points need to be designed with long-term maintainability and extensibility in mind.
- High Blast Radius — errors in foundational code propagate through the entire system, making the cost of an agent mistake disproportionately high.
Identifying what belongs in the Core Nucleus is a key governance decision. The boundary is not fixed; as confidence in agent capabilities and evaluation coverage grows, components may graduate from hand-written Core Nucleus code to agent-generated code with appropriate guardrails.