The AI community is pretty incredible with how much it has worked together and grown over the last couple of years. Its an amazing place to be for learning as there is not only huge amount of amazing libraries and frameworks to look at, the open source communities behind them are amazing and passionate too.
If you’re like me though, you will eventually have to talk to CEOs, boards and Investors about how safe, how secure and how supported these are and how we manage risk around it. In cases this like, Open AI, Google, AWS and Microsoft stand out as trusted partners and are likely to be your go to for using their enterprise supported products and services.
As a long time Microsoft advocate, when I reach this point my preference is always Microsoft. So here I will talk abit about how we can use Microsoft tech to reach our AI goals.
The enterprise buzzword right now is all about autonomous agents—AI systems that can independently complete tasks, make decisions, and manage complex workflows. Microsoft Semantic Kernel provides developers with the tools needed to build these agents by using three core components: Plugins, Planners, and Personas.
So here I will go into the role each of these components plays and discuss how the can bye applied in enterprises to build intelligent, dynamic AI systems.
Note: I could provide my code samples here, but Microsoft has actually provided a really great page in MSLearn, so here I will talk you through the concepts only. Go to this link for up to date code straight from the horses mouth. https://learn.microsoft.com/en-us/semantic-kernel/concepts/agents?pivots=programming-language-csharp
Plugins – Extend your AI with External Skills
Plugins in Semantic Kernel provide the capabilities or “skills” that an agent can use to interact with external systems, services, or data. These plugins serve as the agent’s toolbox, enabling it to perform tasks by calling APIs or accessing external resources. Whether it’s fetching data from a CRM system, sending an email, or checking inventory levels, plugins expand the scope of what the agent can accomplish.
How Plugins Work
Plugins are modular and can be custom-built or integrated from existing systems. When an AI agent receives a task, it checks its available plugins and uses them to achieve the goal. For example, if an agent is tasked with booking a flight, it might use a plugin to interact with your internal support ticket API, check the ticket exists and update it automatically so all customer contact is logged properly.
Planners – Orchestrating Tasks to Achieve Goals
While plugins give the agent its capabilities, Planners determine how to achieve a goal by breaking it down into tasks and orchestrating the necessary steps. A Planner is responsible for the sequencing of tasks, dynamically adjusting the workflow as needed, and calling the right plugins at the right time.
How Planners Work
Planners take high-level goals and map out the best way to accomplish them. For example, if an AI agent is tasked with processing an order, the Planner determines the necessary steps: checking inventory, processing the payment, generating an invoice, and sending a confirmation email. Planners ensure that tasks are executed in the correct order and can adapt to changes in real-time, such as errors or new data.
Its worth highlighting at this point that the early versions of planners, including Stepwise and Handlebar planner have been superseded by leveraging Function Calling as a means of automatically allowing the AI to invoke custom functions as required. Its also worth noting that when writing this, only Open AI models support Function Calling so some careful consideration might be required if you’re building out a solution that does not use Open AIs models.
Personas – Defining How AI Interacts with Users
Personas shapes how the AI interacts with users or other systems by defining its communication style, tone, and behavioral patterns. Personas allow the AI to adapt its behavior based on the context of the interaction, making it feel more natural and appropriate for different scenarios.
How Personas Work
Personas are predefined roles that dictate how the AI responds to users or other systems. For example, an AI agent with a “technical support” persona might use technical language when interacting with engineers but switch to simpler explanations for non-technical users. This adaptability makes interactions more relevant and human-like, allowing the AI to better serve different types of users.
Combining Plugins, Planners, and Personas to Build Autonomous Agents
While each of these components—Plugins, Planners, and Personas—plays a critical role individually, they work best when combined to create a fully autonomous AI agent. Together, they allow AI systems to not only perform complex tasks but also do so in a way that is adaptive, goal-oriented, and user-focused.
How These Components Work Together
Let’s look at an example that shows how these elements come together:
- Plugins: The agent uses plugins to interact with external systems, such as sending emails, querying databases, or booking appointments.
- Planners: The Planner handles the orchestration of tasks, deciding the order of actions needed to complete the goal, such as scheduling a meeting after checking availability.
- Personas: The Persona shapes how the AI interacts with different users, ensuring that the tone and communication style match the context—whether it’s formal, casual, or technical.
Enterprise Use Cases
In an enterprise setting, the combination of these components can revolutionize automation and decision-making:
- Customer Service Automation: Imagine an AI agent in a customer service department that can handle a full range of inquiries. The Planner would orchestrate the entire workflow—from retrieving customer data with a plugin to suggesting solutions based on previous interactions, all while using a Persona tailored to customer support to maintain a friendly and empathetic tone. The agent could escalate to human support only when necessary, reducing workloads while improving response times.
- Business Process Automation: In finance or HR, an AI agent could automate multi-step processes like invoice approval, onboarding, or payroll processing. The Planner would sequence tasks such as verifying documents or processing payments, while Plugins would connect to ERP or HR systems to retrieve or update data. The Persona would adapt to different departments, using a formal tone for executives and a more casual tone for general staff interactions.
- Sales and Marketing: For sales teams, an AI agent could guide customers through the sales funnel autonomously. It would use Plugins to access CRM data and track customer history, and the Planner would determine the best course of action, such as sending personalized follow-ups or arranging product demos. The Persona would be sales-oriented, using persuasive language to build rapport and close deals.
Summary
Microsoft Semantic Kernel gives us a solid framework for building autonomous AI agents using nd its three key components, Plugins, Planners, and Personas, each play a critical role in shaping how these agents operate. By using the Semantic Kernel enterprises can create intelligent, goal-oriented systems that adapt to both user needs and complex workflows in real time.
Masters Development 