Modern organizations need intelligent systems that go beyond basic automation by perceiving, reasoning, and acting independently. Agentic AI fulfills these requirements. Microsoft’s Azure AI Foundry provides tools and templates for building advanced agents that can achieve goals, adapt to change, and integrate with other systems, supporting use cases like operations, security, and personalized training. 

Image from Microsoft Learn 

To further support innovation, Microsoft provides a comprehensive Learning plan titled Create Agentic AI solutions by using Azure AI Foundry. This resource delivers detailed step-by-step guidance for building, deploying, and optimizing agentic AI, simplifying the transition from concept to fully functional autonomous solutions. The plan is available at. The image below illustrates the process of adding data to Azure AI Foundry. https://aka.ms/CreateAgenticAISolutions 

 

Image from Microsoft Learn Why Agentic AI Is a Game-Changer 

Traditional automation relies on rigid rules and workflows. But today’s challenges, dynamic threats, diverse user needs, and complex decision-making, require systems that can think and act. 

Agentic AI shifts the focus from task execution to goal achievement. These agents: 

• Understand context and adapt over time 

• Use memory to learn from past interactions 

• Collaborating with other agents and tools 

• Respond to natural language prompts 

 

Image from Microsoft Learn 

Key Capabilities of Agents Built on Azure AI Foundry 

Azure AI Foundry offers a robust framework for building agentic AI solutions: 

• Goal-Oriented Design: Agents operate based on clearly defined objectives and constraints, ensuring their actions consistently align with desired outcomes. 

• Thread-Based Memory: Agents maintain context within session-specific threads. To preserve memory across sessions, especially when triggered via webhooks, developers must manually persist and reuse the Thread ID, as long-term memory isn’t built-in yet. 

• Tool Use: Agents can interact with external APIs, databases, and services, giving them the flexibility to access information and perform specialized tasks. 

• Multi-Agent Collaboration: Agents are capable of working together, sharing information, and coordinating strategies to tackle complex, multi-faceted challenges. 

These capabilities allow organizations to build agents that are not only intelligent but also scalable and adaptable. 

 How It Works 

Azure AI Foundry streamlines setup and deployment processes; however, building agents still requires some development skills. The fundamental intelligence and behavior of each agent are determined by custom code written by developers. Creating an agent in Azure AI Foundry follows a structured but adaptable process: 

1. Define the goal. What should the agent achieve? 
2. Choose a template. Start with a pre-built agent or build your own. 
3. Configure memory and tools. To enable persistent memory, customers must manually set this up using an open-source memory tool or third-party services. Additionally, connect external services as needed. 
4. Deploy via Foundry SDK or Foundry portal. Launch your agent in a secure, scalable environment. 
5. Monitor and iterate. Use dashboards and logs to refine agent behavior over time. 

Explore the https://learn.microsoft.com/en-us/azure/ai-foundry/agents/quickstart?pivots=ai-foundry-portal for hands-on steps 

Advanced Model Fine-Tuning Techniques 

To unlock the full potential of agentic AI, Azure AI Foundry empowers developers and trainers to apply advanced model fine-tuning techniques that tailor large language models (LLMs) to specific domains, tasks, and organizational goals. These methods enhance model performance, improve instruction, and ensure alignment with compliance and safety standards. Fine-tuning customizes pre-trained models for your specific domain and tasks, improving accuracy, reducing costs, and enabling faster, more relevant responses. Azure AI Foundry supports a variety of fine-tuning strategies: 

• Supervised Fine-Tuning (SFT) – Aligns models with structured datasets and enterprise tone-of-voice. 

• Reinforcement Fine-Tuning (RFT) – Enables iterative learning for complex decision-making. 

• Direct Preference Optimization (DPO) – Improves model behavior based on user preferences. 

 

Use supervised fine-tuning for tone and content, and reinforcement fine-tuning for reasoning, distillation for cost reduction. Azure AI Foundry streamlines fine-tuning with built-in safety, global training, and cost-effective developer tools.  

Follow a practical workflow when fine-tuning with these 5 phases: 

1. Domain Specialization 
2. Task Performance: 
3. Style and Tone 
4. Instruction Following 
5. Compliance and Safety 
6. Language or Cultural Adaptation 

This structured approach ensures that agents are not only intelligent and scalable but also deeply aligned with enterprise needs. Learn more Top use cases for fine-tuning 

Real-World Use Cases 

Security agents identify potential threats and suggest policy adjustments. 

Training agents create personalized learning plans for technical teams. 

Translation agents, such as the Teams Interpreter Agent, facilitate real-time interpretation. 

Planning agents compile updates and assist in decision-making processes. 

Use Cases in Action 

Security automation enables organizations to identify emerging threats, isolate compromised systems, and recommend necessary policy modifications. In training optimization, learners receive individualized content tailored specifically to their roles and skill levels, promoting more effective upskilling. Language translation agents facilitate real-time interpretation during conversations, helping organizations overcome communication challenges. Portfolio planning agents compile updates, analyze trends, and assist in data-driven decision-making for various business functions. 

Domain-specific fine-tuning allows agents to be adapted to fields such as healthcare, finance, or law by utilizing techniques like Supervised Fine-Tuning (SFT). Instructional alignment enhances agent compliance with specified tones, formats, and multi-step logic through methods such as Direct Preference Optimization (DPO) and Reinforcement from Feedback Tuning (RFT). Compliance and safety requirements are integrated into model behavior by means of targeted fine-tuning, ensuring that regulatory standards are met. 

Several common fine-tuning challenges can arise. For instance, limited labeled data for niche domains can be addressed using few-shot learning with domain-specific prompts, reducing dependency on large datasets. Overfitting during fine-tuning is mitigated by applying regularization techniques and validating results with cross-domain test sets. High compute costs are controlled by employing parameter-efficient methods or leveraging Azure’s managed compute environments. To prevent misalignment with enterprise tone or compliance requirements, organizations can incorporate reinforcement learning with human feedback, such as RLHF or DPO, to guide model behavior effectively. 

Learn more here: 

Fine-tune models with Azure AI Foundry – Overview of fine-tuning strategies and use cases. 

Customize a model with fine-tuning – Step-by-step guide to fine-tuning in Azure AI Foundry. 

AI model fine-tuning concepts – Foundational concepts and transfer learning.  

Augment LLMs with RAGs or Fine-Tuning – Comparison of RAG and fine-tuning for enterprise applications. 

Requirements 

To get started, you’ll need: 

• An Azure subscription 

• Access to Azure AI Foundry and Azure AI Foundry 

• Optional integrations with Azure OpenAI, Cognitive Services, and plugins 

 

Pros and Cons 

Call To Action 

First read the clear learning path is provided with sequential steps for building, deploying, and refining agentic AI. 

Create Agentic AI solutions by using Azure AI Foundry https://aka.ms/CreateAgenticAISolutions 

Next, you should be ready to build your first agent! 

1. Visit https://aka.ms/CreateAgenticAISolutions 
2. Explore the https://learn.microsoft.com/en-us/azure/ai-foundry/agents/quickstart?pivots=ai-foundry-portal  
3. Choose a template and define your agent’s goals 
4. Configure memory, tools, and reasoning 
5. Deploy and monitor agent performance 

 

About the Author 

Hi! I’m Jacques “Jack”, Microsoft Technical Trainer. I help learners and organizations adopt intelligent automation through Microsoft technologies. This blog reflects my experience guiding teams in building agentic AI solutions that are not only powerful but also intuitive and scalable. With Azure AI Foundry, we’re entering a new era of intelligent systems, one where agents don’t just follow instructions, they pursue goals. 

#MicrosoftLearn #SkilledByMTT