- Forrester Best Practice Report
The Architect’s Guide to Agentic AI
Co-Inventing the Next Era of Software
Agentic AI is moving quickly from experimentation to enterprise deployment. But building an agent that works is only part of the challenge.
Enterprise teams also need to design for scale, security, governance, resilience, and flexibility - without creating another layer of architectural complexity.
In this report, Forrester explores the core components and emerging architectural patterns behind agentic AI systems to help architects make better design decisions as they move from proof of concept to production.
What does it take to architect agentic AI for the enterprise?
There is no single architecture for every agentic system.
The right approach depends on what the agent needs to accomplish, how much autonomy it requires, which tools and data it can access, and how multiple agents need to work together.
Forrester examines the building blocks that enterprises should consider together rather than in isolation.
In this report, you’ll learn about:
- Agentic runtimes - Build the foundation for flexible and resilient agents.
- Reasoning - Understand how reasoning components work with foundation models.
- Memory - Explore the evolving role of memory in agentic applications.
- Tools - Enable agents to discover and safely use the capabilities required to take business actions.
- Guardrails - Control agent behavior using code, foundation models, and business rules.
- Security - Prepare for the new attack surfaces and risks introduced by agentic systems.
- Testing & Evaluation - Treat evaluation as an architecture-wide requirement rather than an afterthought.
- Orchestration - Coordinate agents, models, tools, memory, and workflows into a cohesive system.
Choose the right architecture for the job
The report also explores different architectural approaches for agentic systems, including:
- Single-agent architectures for targeted products and workflows
- Sequential multiagent architectures for linear task flows
- Orchestrated multiagent architectures for assigning, distributing, and assembling work
- Collaborative multiagent patterns for more complex agent-to-agent interaction
Understanding when to use each approach can help teams avoid unnecessary complexity while designing systems that can evolve as requirements change.
Move from agent experimentation to enterprise architecture
As agentic AI becomes part of enterprise software, architecture decisions made today will determine how effectively these systems can be governed, secured, tested, and scaled tomorrow.
Get the Forrester report to explore the architectural components and patterns shaping the next generation of agentic applications.