Agent Design · Solution Patterns

Agentic Solution Patterns

Enterprise agentic solutions exist on a spectrum of complexity. Four patterns — from simple chatbots to multi-agent systems — define the landscape, and production systems need both deterministic and agentic processing.

The Fundamental Tension

Deterministic Processing

Predictable, repeatable execution with defined paths — the foundation of current enterprise software.

Repeatable and predictable, but rigid
Abundant in the enterprise
Basis for current software engineering
You need
both

Agentic Processing

Dynamic, probabilistic execution that adapts to context — capable of self-improvement but expensive.

Dynamic and adaptable, but expensive
Capable of self-improvement
Best practices still being determined
Integration Directions
Combining Deterministic and Agentic Processing
Two complementary patterns for integrating agents with existing enterprise systems.
calls

Agent Calls Processes

An agent uses deterministic workflows as tools — invoking existing business processes, integrations, and APIs to accomplish tasks reliably.

calls

Process Calls Agents

A deterministic workflow delegates specific steps to agents — adding intelligence at decision points within an otherwise structured pipeline.

Agent Tool Types

Knowledge Tools

Read-only retrieval. Query databases, search knowledge bases, look up information without modifying state.

Non-destructive

Action Tools

State-changing operations. Create records, process transactions, send messages, modify external systems.

Destructive

Composite Tools

Orchestrate multiple operations. Combine knowledge retrieval and actions into coordinated multi-step tool sequences.

Mixed
The Spectrum
Four Agentic Solution Patterns
From simple conversational interfaces to fully collaborative multi-agent systems — each pattern adds complexity, capability, and enterprise value.
Lower complexity Higher complexity
Pattern 1

Simple Chatbots

Standalone agents for Q&A with human users. Conversational interfaces backed by knowledge retrieval.

Elements
User Agent Knowledge Tools
Enterprise value
Low
Architecture
Pattern 2

Worker Agents

Functional agents that use both knowledge and action tools to perform tasks on behalf of users.

Elements
User Agent Knowledge Tools Action Tools
Enterprise value
Medium
Architecture
Pattern 3

Agentic Workflows

Deterministic workflows that combine agents and other software components to execute business processes.

Elements
User Agent Knowledge Tools Action Tools Orchestrator Events
Enterprise value
High
Architecture
Pattern 4

Multi-Agent Systems

Multiple specialized agents collaborating in shared conversation spaces to handle complex, cross-domain tasks.

Elements
User Agents Knowledge Tools Action Tools Orchestrator Events
Enterprise value
Highest
Architecture
User Conversation Agent Retrieval Knowledge Base
Pattern 1 Architecture

Simple Chatbots

A linear flow: the user converses with a single agent, which retrieves information from a knowledge base. No state is modified, no external actions are taken. The agent is purely conversational.

Example
An internal support chatbot that answers employee questions by searching the company knowledge base.
User Conversation Agent Retrieval Knowledge Tools Execute Action Tools Read/Write Data Store
Pattern 2 Architecture

Worker Agents

A hub-and-spoke pattern: the agent sits at the center, with connections radiating out to knowledge tools, action tools, and data stores. The agent reasons about which tools to invoke and in what order.

Example
A healthcare insights advisor that retrieves patient data, processes queries, and delegates tasks across multiple backend systems.
User Agent Workflow Orchestrator Knowledge Tools Action Tools Event Stream ERP POS CRM
Pattern 3 Architecture

Agentic Workflows

A hybrid architecture: a deterministic workflow orchestrator controls the overall process, while agents handle steps that require reasoning. Event streams connect the components. Enterprise systems (ERP, POS, CRM) are accessed through governed tool interfaces.

Example
A retail inventory system where a workflow orchestrator coordinates stock analysis, demand prediction, and reorder execution — with an agent providing intelligent decision-making at key steps.
Customer Request CSR Agent Conversation Space Repair Tech Parts
Pattern 4 Architecture

Multi-Agent Systems

A collaborative architecture: multiple specialized agents communicate through a shared conversation space. The customer interacts with a CSR agent, which collaborates with repair and parts specialists. Each agent brings domain-specific tools and knowledge, coordinating dynamically to resolve complex requests.

Example
A field service system where a CSR agent, repair technician agent, and parts specialist agent collaborate in real time to diagnose issues, check parts availability, and schedule repairs.