Agent Factory: The new era of agentic AI—common use cases and design patterns

Agents don’t just share information; they think, act, and work together to connect knowledge with real results. Discover more about agentic AI in Azure AI Foundry.

This blog post kicks off a six-part series called Agent Factory, where we’ll explore best practices, design strategies, and tools to aid you in adopting and developing agentic AI.

Why Enterprises Need Agentic AI Beyond Just Knowledge

Retrieval-augmented generation (RAG) has revolutionised enterprise AI, allowing teams to quickly discover insights and get answers like never before. For many, it served as a springboard into advanced tools like copilots and chatbots, which have made support easier and cut down the time needed to find information.

However, simply providing answers often doesn’t lead to significant business results. Most enterprise tasks require actions, such as submitting forms, updating records, or managing complex systems. Conventional automation tools—like scripts, Robotic Process Automation (RPA) bots, and manual transitions—struggle to adapt to change and scale, leaving teams dealing with frustrations and inefficiencies.

This is where agentic AI plays a transformative role. These agents not only deliver information; they also think, act, and collaborate, effectively filling the gap between knowledge and actionable results. This ushers in a new age of enterprise automation.

Key Patterns of Agentic AI: Foundations for Enterprise Automation

The transition from merely obtaining information to taking real-world actions starts with agents that utilise tools, but enterprise needs extend further. Effective automation hinges on agents that can reflect on their tasks, plan elaborate processes, collaborate across different areas, and adjust to changes in real-time, rather than merely executing simple tasks.

Below are five essential patterns that are already paving the way for automation in the industry. These building blocks can be combined to unlock significant advances in automation.

1. Tool Use Pattern: From Advisor to Operator

Today’s agents shine by generating real outcomes. They engage directly with enterprise systems—pulling data, utilising Application Programming Interfaces (APIs), initiating workflows, and conducting transactions. These agents are now not only offering answers but also carrying out tasks, updating records, and managing workflows from start to finish.

Fujitsu has revamped its sales proposal process with dedicated agents for data analysis, market research, and document generation—each using specific APIs and tools. Instead of just answering “what should we pitch,” these agents create complete proposal packages, reducing production time by a remarkable 67%.

2. Reflection Pattern: Self-Improvement for Reliability

Once agents can take action, the next crucial step is reflection—allowing them to evaluate and enhance their outputs. This ability helps agents identify mistakes and improve quality without always relying on human input.

In critical fields like compliance and finance, even a minor error can be very costly. Through internal checks and feedback loops, agents can automatically correct omissions, verify calculations, or ensure messages comply with guidelines. Even coding assistants, like GitHub Copilot, depend on rigorous internal testing before finalising outputs. This self-optimisation cycle minimises errors and ensures that AI-driven processes remain safe, reliable, and capable of being audited.

3. Planning Pattern: Breaking Down Complexity for Robustness

Most organisational processes involve multiple steps and complications, not simple linear tasks. Planning agents tackle this by decomposing high-level objectives into smaller, actionable items, monitoring progress, and adapting as circumstances change.

ContraForce’s Agentic Security Delivery Platform (ASDP) has streamlined its partner’s security service with planning agents that break down incidents into phases like intake, impact assessment, execution, and escalation. After completing each phase, the agent checks for follow-ups, ensuring nothing is overlooked. The outcome? An impressive 80% of incident investigation and response is now automated, with each full investigation costing less than $1.

Planning often blends tool use and reflection, demonstrating how these patterns support one another. A key benefit is adaptability: plans can be dynamically formulated by a large language model (LLM) or follow a specific sequence, based on what’s needed.

4. Multi-Agent Pattern: Collaboration at Machine Speed

No single agent can accomplish everything. Enterprises generate value through collaborative teams, and the multi-agent pattern reflects this by interlinking networks of specialist agents—each dedicated to different parts of a workflow—under one orchestrator. This modular structure promotes flexibility, scalability, and smooth evolution while maintaining clear governance and responsibilities.

Contemporary multi-agent setups utilise various orchestration methodologies, often combined to meet genuine enterprise needs. These methodologies can be either driven by LLMs or deterministic: sequential orchestration (where agents refine a document progressively), concurrent orchestration (agents operate simultaneously and merge results), group chat/maker-checker (agents debate and validate outputs collaboratively), dynamic handoff (real-time triage or routing), and magentic orchestration (a supervisory agent oversees all tasks until completion).

JM Family has embraced this model with its business analyst/quality assurance (BAQA) Genie, deploying agents focused on requirements, story crafting, coding, documentation, and Quality Assurance (QA). By coordinating through an orchestrator, they managed to streamline development cycles—cutting the time for requirements gathering and test design from weeks to mere days, saving up to 60% of QA time in the process.

5. ReAct (Reason + Act) Pattern: Adaptive Problem Solving in Real Time

The ReAct pattern enables agents to tackle issues live, especially in situations where static plans are ineffective. Instead of sticking to a strict script, ReAct agents alternate between reasoning and acting—taking a step, observing the results, and deciding their next move. This allows them to adjust to uncertainties, changing requirements, and scenarios where the best course of action isn’t clear-cut.

For instance, in enterprise IT support, a virtual agent utilising the ReAct pattern can identify problems on the spot: it can pose clarifying queries, inspect system logs, test potential solutions, and modify its approach as new data surfaces. If the issue escalates or falls outside its expertise, the agent can refer the case to a human expert, complete with a concise summary of its actions.

These patterns can be integrated. The most effective agentic solutions combine tool use, reflection, planning, multi-agent interaction, and responsive reasoning—leading to automation that’s quicker, smarter, safer, and ready for real-world applications.

The Importance of a Unified Agent Platform

Creating intelligent agents involves much more than just initiating a language model. When transitioning from concept to practical application, teams often encounter various challenges:

• How can I effectively link multiple steps together?
• How do I ensure agents have secure and responsible access to business data?
• How can I monitor, assess, and enhance agent performance?
• How can I maintain security and identify different components within agents?
• How can I scale from a single agent to a network of agents or connect to others effectively?

Many teams find themselves creating custom structures—DIY orchestrators, logging mechanisms, tool managers, and access controls. This not only lengthens the time-to-value but also increases risks and leads to fragile solutions.

This is where Azure AI Foundry steps up—not merely as a collection of tools but as a comprehensive platform aimed at guiding agents from concept to enterprise-level implementation.

Azure AI Foundry: Unified, Scalable, and Built for Real-World Needs

Azure AI Foundry is engineered specifically for this new era of agentic automation. It offers a single, comprehensive platform tailored to meet the requirements of both developers and enterprises, merging rapid innovation with robust enterprise-level controls.

With Azure AI Foundry, teams can:

• Prototype Locally, Scale Up: Develop and test agents locally, then effortlessly transition to a cloud-based runtime without needing to rewrite. Take a look at how to get started with Azure AI Foundry SDK.
• Choose Flexible Models: Select from Azure OpenAI, xAI Grok, Mistral, Meta, and over 10,000 open-source models—all accessible through a unified API. A Model Router and Leaderboard assist in selecting the most suitable model, balancing performance against cost and expertise. Explore the Azure AI Foundry Models catalog.
• Create Modular Multi-Agent Architectures: Link specialised agents and workflows, reusing frameworks across teams. Check out how to use connected agents in Azure AI Foundry.
• Instant Integration with Enterprise Systems: Make use of over 1,400 built-in connectors for SharePoint, SaaS, and other business applications, complete with native security and policy support. Discover what tools are available in Azure AI Foundry.
• Support Openness and Interoperability: With native support for open protocols like Agent-to-Agent (A2A) and Model Context Protocol (MCP), your agents can operate across multiple clouds and platforms. Learn how to connect to a Model Context Protocol Server Endpoint in Azure AI Foundry.
• Enterprise-Level Security: Each agent is assigned a managed Entra Agent ID, robust Role-based Access Control (RBAC), On Behalf Of authentication, and policy enforcement—ensuring that only the right agents access the necessary resources. Check out how to implement a virtual network with Azure AI Foundry.
• Comprehensive Observability: Get in-depth visibility with step-level tracing, automated evaluation, and integration with Azure Monitor—supporting compliance and continuous improvement at scale. Read more about how to monitor Azure AI Foundry Agent Service.

Azure AI Foundry isn’t just a set of tools; it’s the cornerstone for creating secure, scalable, and intelligent agents throughout the modern enterprise.
This is how organisations transition from isolated automation to genuine, comprehensive business transformation.

Stay tuned: In the upcoming entries of our Agent Factory series, we’ll demonstrate how to implement these principles—illustrating how to create secure, coordinated, and interoperable agents using Azure AI Foundry, from initial development to enterprise deployment.