Prompt Systems and Agent Orchestration

Volume two of The Full-Stack AI Engineering Series, covering the application layer at which prompts stop being text files and become governed agents. Twelve chapters develop AgentMesh, a fictional catalog of record operated by the regulated fintech Nebula Financial, and treat agent taxonomy in financial services, the Planner, Executor, Verifier, Generator pattern that opens a single agent into four named roles, the training of planners from live execution trajectories, workflows modeled as directed graphs that can be inspected and refined, agent registries and discovery protocols, human review designed as a runtime orchestration layer with tiers mapped to risk, reviewer feedback structured as first-class data, and the safety boundaries that separate an agent permitted only to read from one permitted to act. A case study carries automated regulatory disclosure drafting end to end, and a closing chapter assembles the blueprint and a design-patterns catalog. Specifications are conceptual, presenting schemas, policy objects, trace shapes and architecture diagrams rather than production code. Addressed to AI engineers, application developers and agent builders working under regulatory constraint.

שנה

2026

שפה

English

סדרה

The Full-Stack AI Engineering Series, volume 2

הוצאה

iSystematic Inc.

מהדורות

1

עמודים

260

ISBN

9781067831769

מקור הנתונים: Wikidata