Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Architectures for Building Agentic AI

Sławomir Nowaczyk

TL;DR

The paper argues that reliable agentic and generative AI systems are primarily a matter of architecture, not just model quality. It presents a principled, componentized blueprint—covering goal management, planning, tool routing, execution, memory, verification, safety monitoring, and telemetry—with disciplined interfaces and assurance loops to bound risk. A taxonomy of modern architectures (tool-using, memory-augmented, planning/self-improvement, multi-agent, embodied/web) is analyzed for how each pattern shapes reliability and failure modes, illustrated by a running diagnosis agent. Practical design guidelines (typed schemas, idempotency, permissioning, transactional semantics, memory hygiene, runtime governance, and simulate-before-actuate) provide templates for auditable, safe autonomous systems and set the stage for data coverage, confidence estimation, cybersecurity, monitoring, and governance in subsequent chapters.

Abstract

This chapter argues that the reliability of agentic and generative AI is chiefly an architectural property. We define agentic systems as goal-directed, tool-using decision makers operating in closed loops, and show how reliability emerges from principled componentisation (goal manager, planner, tool-router, executor, memory, verifiers, safety monitor, telemetry), disciplined interfaces (schema-constrained, validated, least-privilege tool calls), and explicit control and assurance loops. Building on classical foundations, we propose a practical taxonomy-tool-using agents, memory-augmented agents, planning and self-improvement agents, multi-agent systems, and embodied or web agents - and analyse how each pattern reshapes the reliability envelope and failure modes. We distil design guidance on typed schemas, idempotency, permissioning, transactional semantics, memory provenance and hygiene, runtime governance (budgets, termination conditions), and simulate-before-actuate safeguards.

Architectures for Building Agentic AI

TL;DR

The paper argues that reliable agentic and generative AI systems are primarily a matter of architecture, not just model quality. It presents a principled, componentized blueprint—covering goal management, planning, tool routing, execution, memory, verification, safety monitoring, and telemetry—with disciplined interfaces and assurance loops to bound risk. A taxonomy of modern architectures (tool-using, memory-augmented, planning/self-improvement, multi-agent, embodied/web) is analyzed for how each pattern shapes reliability and failure modes, illustrated by a running diagnosis agent. Practical design guidelines (typed schemas, idempotency, permissioning, transactional semantics, memory hygiene, runtime governance, and simulate-before-actuate) provide templates for auditable, safe autonomous systems and set the stage for data coverage, confidence estimation, cybersecurity, monitoring, and governance in subsequent chapters.

Abstract

This chapter argues that the reliability of agentic and generative AI is chiefly an architectural property. We define agentic systems as goal-directed, tool-using decision makers operating in closed loops, and show how reliability emerges from principled componentisation (goal manager, planner, tool-router, executor, memory, verifiers, safety monitor, telemetry), disciplined interfaces (schema-constrained, validated, least-privilege tool calls), and explicit control and assurance loops. Building on classical foundations, we propose a practical taxonomy-tool-using agents, memory-augmented agents, planning and self-improvement agents, multi-agent systems, and embodied or web agents - and analyse how each pattern reshapes the reliability envelope and failure modes. We distil design guidance on typed schemas, idempotency, permissioning, transactional semantics, memory provenance and hygiene, runtime governance (budgets, termination conditions), and simulate-before-actuate safeguards.

Paper Structure

This paper contains 12 sections, 2 figures.

Table of Contents

  1. Introduction: purpose, scope, and architecture reliability
  2. Classical foundations
  3. Reactive, deliberative, and hybrid architectures
  4. Belief–Desire–Intention to structure behaviour
  5. A taxonomy of modern agentic architectures
  6. Tool-using agents
  7. Memory-augmented agents
  8. Planning- and self-improvement agents
  9. Multi-agent systems
  10. Embodied or web agents
  11. Architectural building blocks and interfaces
  12. Summary

Figures (2)

  • Figure 1: Reliability is earned architecturally via typed interfaces, least privilege, assurance hooks (validators, simulation), budgets/termination, safety, and observability.
  • Figure 2: Comparison of agentic architecture families across core design facets, showing how each approach shapes capabilities, risks, and the reliability tools required to control behaviour.