← Back to guides
JevMade field notes / Research writeup
Separating Agent decisions from LLM generation in system design
A layered agent architecture assigns bounded judgments to Jev, prose and planning to generative models, and effects to deterministic code.
Original by daryl / Jev TutorialAgent workflowsArticleOriginal published Source reviewed
Before you dive in
What you’ll find in the original
- Use a fast decision layer only where the output space is defined in advance.
- Keep permissions, side effects, and rollback in deterministic application code.
- Route high-risk or low-confidence cases to a person or stronger model.
Worth knowingThe page says it adapts and independently restructures an article by daryl at Tencent Technology Engineering.