12-Factor Agents: Patterns of Great LLM Applications | Hacker News

Could they use this model—they would dub it the “pseudo-count”—to motivate an agent to seek these novel states?

1. Delegation : Use cheaper, faster models for cost and speed gains.


For example, Opus can delegate to Haiku to read a book and return relevant passages. This works well if the task description & result are more compact than the full context.
2. Parallelization : Cut latency (but

mimicking human-like intelligence but at a transformative scale and speed.

Think of agents as AI employees rather than contractors: they maintain context, have specific roles, and can use tools to accomplish tasks.