Understand AI agents and Microsoft Foundry Agent Service
Unit 2 of 10 · Module 7: Develop AI agents with Microsoft Foundry and Visual Studio Code
Go deeper
Unlocks when you've watched the lesson.
Read this lesson
What agents are
An AI agent is a piece of software that uses generative AI to understand a task and carry it out, for a person or for another program. A traditional app is like a rail cart that can only follow its track; an agent is more like a robot crew member that reads the situation, makes decisions, and acts on its own to reach a goal. It pairs a capable AI model with specialized tools, so the automation can adapt as situations change. Agents help in four big ways. They take over routine work, so people can spend their time on strategy and creative problems. They sharpen decisions: where a plain chat model only produces text, an agent can dig into data, spot trends, predict what's next, and recommend an action based on live information. They let an operation grow without needing more people or money at the same rate. And, like any software, they never clock off.
Where agents work
You'll find agents at work in almost every industry. Personal productivity agents book meetings, send emails, and keep to-do lists tidy; Microsoft 365 Copilot, for example, helps write documents, build presentations, and make sense of data. Research agents watch trends, collect data, and write up reports for teams in finance, healthcare, and marketing. Sales agents look into leads and qualify them, send tailored follow-ups, and set up calls, leaving salespeople free to close deals. Customer service agents deal with everyday questions and common problems; the cinema chain Cineplex, for instance, uses one to handle refund requests, which cut handling time and made customers happier. And developer agents, like GitHub Copilot, help review code, fix bugs, and manage repositories.
Security by design
Because agents touch sensitive data, make decisions, and act without asking, security has to be designed in from day one, like the hull of a space station. The main threats: data leakage, where an agent reveals confidential information; prompt injection, where hidden instructions trick it into leaking data or doing things it isn't allowed to; and unauthorized access or privilege escalation, like an agent running admin actions in a CRM. There are more: data poisoning, where bad data leads to unsafe choices; a weak link in the supply chain, like a plugin or API the agent relies on; too much trust in its autonomy, like an agent sending a payment by mistake; thin logging, so misuse goes unnoticed; and model inversion, where repeated questions coax out private training data. To lower the risk, give each agent only the access its role needs, filter and validate prompts, require a human to approve sensitive actions, log and trace everything an agent does, check third-party dependencies regularly, and keep retraining and validating models to catch drift or poisoning.
Foundry Agent Service
Microsoft Foundry Agent Service is a fully managed service for building, deploying, and scaling agents securely, with no compute or storage to look after: mission control for your robot crew. You shape agents with custom instructions and powerful tools, either in the Foundry portal, or from your own app in fewer than fifty lines of code. It supports two main kinds of agents. Declarative agents are defined by configuration: a prompt agent is a single agent with a model, instructions, and tools, the most common kind and the one this module focuses on; you may also meet workflow agents, teams described in YAML, but Foundry retires them on December first, twenty twenty-six. Hosted agents are agents you build in code, shipped as a container or as your code, while the platform runs the infrastructure for you. Its key features: automatic tool calling from start to finish; conversation state kept safe for you through the Responses API; a big catalog of built-in and community tools; a choice of models; enterprise-grade security, with keyless sign-in and content safety filters; storage managed by the platform, or your own Azure Blob storage; and built-in observability and tracing.
Facts from Microsoft's documentation: What is Microsoft Foundry Agent Service? (CC BY 4.0), rewritten in Loop's own words and examples. Narration, animation, and quiz by Loop. Microsoft, Azure, and Microsoft Foundry are trademarks of Microsoft. Loop is independent and not affiliated with or endorsed by Microsoft.