Why Not LLM Wrappers?

Compare LLM wrapper agents to Artemis City’s kernel-first architecture and learn why separating governance from inference improves reliability.

Wrapper-first systems are useful for prototyping but weak for long-lived operations.

Common Failure Modes

• Non-deterministic tool routing
• Context and memory drift over long tasks
• Limited accountability for runtime decisions
• High cost from repeated full-inference planning loops

Kernel-First Alternative

Artemis City separates runtime control from inference:

• Kernel: governance, routing, state transitions
• Agents/LLMs: task execution within boundaries

This keeps behavior predictable while still using modern model capabilities.