How To.. Build agents for, Agent Builder, Copilot Studio, and Foundry

Introduction

Navigating Microsoft's agent-building landscape can be complex. This guide breaks down the three primary paths—Agent Builder, Copilot Studio, and Foundry—to help you distinguish between no-code, low-code, and full-code approaches. Whether you need rapid personal automation or enterprise-grade workflows, this article helps you choose the right tool for your workload.

Getting Started

It’s no secret that artificial intelligence is reshaping how we work. Whether you are automating mundane tasks or developing intelligent solutions, if you’re operating inside the Microsoft ecosystem, you’ve probably noticed there are three distinct paths to creating agents. These paths certainly overlap, but they’re not interchangeable—each one sits at a completely different layer of the stack. Let's break down the 'what', 'where', 'why', and 'how' of Microsoft’s agent-building avenues to help you find the right fit for your workload.

Agent Builder — No‑Code, Chat‑Native, Fastest Path to Value

• Where it lives: Inside Copilot Chat

• Skill level: No code

• Execution model: On‑demand, user‑triggered

• Scheduling: ❌ Not supported

• Best for: Lightweight workflows, personal automations, rapid prototyping

Technical characteristics:

• Built directly inside Copilot using natural language prompts.

• Uses declarative instructions and tool invocation behind the scenes.

• Can call connectors, retrieve data, and run logic, but strictly within a constrained sandbox.

• No branching logic, multi‑step orchestration, or event triggers available.

• Runs synchronously only when directly invoked by the user.

• Great for “micro‑automations” that live close to your daily workflow.

How to Deploy

1. Open the Copilot Chat interface in your web browser.

2. Select 'Create' to launch the Agent Builder configuration pane.

3. Provide a clear natural language prompt to define the agent's behavior and purpose.

4. Add any necessary connectors or data sources within the setup wizard.

5. Click 'Publish' to save your agent; it will immediately become available across your Microsoft 365 apps like Teams and SharePoint.

My real‑world use: I use Agent Builder to spin up exam simulators when I’m prepping for new certifications, or to summarise Teams meetings without the hassle of building a full-blown workflow. It is unequivocally the fastest way to get something genuinely useful up and running.

Copilot Studio — Low‑Code, Orchestration‑Ready, Schedule‑Capable

• Where it lives: Copilot Agent web app

• Skill level: Low‑code

• Execution model: Event‑driven or scheduled

• Scheduling: ✔️ Supported

• Best for: Multi‑step workflows, decision trees, enterprise‑grade automations

Technical characteristics:

• Think of it as “Power Automate meets conversational AI.”

• You design agents using visual flows, conditions, branching, and connectors.

• It fully supports scheduled runs, triggered runs, and manual runs.

• It integrates seamlessly with enterprise systems like Dataverse, Microsoft Graph, and various Azure services.

• It allows for far more complex logic than Agent Builder, including loops, conditions, and variables.

• It supports robust testing, versioning, and publishing lifecycles.

How to Deploy

1. Navigate to the Copilot Studio web app (copilotstudio.microsoft.com).

2. Create a new agent project and open the visual canvas.

3. Drag and drop triggers, actions, and conditional branches to build your workflow.

4. Use the 'Test' pane to validate logic and variables before finalizing.

5. Publish your agent to your designated environment.

6. To deploy, go to the Microsoft 365 admin center, navigate to 'Agents' > 'All agents', select your agent, and choose 'Deploy'.

7. Assign the agent to specific users, groups, or the entire organization as required.

Why it matters: This is where you go when you need repeatable, auditable, enterprise‑friendly automation. It’s the perfect middle ground between a “quick hack” and writing “full code.” I’m still actively learning the nuances of this one myself—it is incredibly powerful, but it certainly requires a different mental model to master.

Foundry — Full‑Code, Python‑Driven, Maximum Flexibility

• Where it lives: Developer‑centric environment

• Skill level: Python / Engineering

• Execution model: Code‑driven, API‑integrated

• Scheduling: ✔️ Supported via orchestrators or external schedulers

• Best for: Custom logic, advanced data processing, AI‑heavy workloads

Technical characteristics:

• Write Python to explicitly define your agent's behaviour, logic, and integrations.

• It fully supports custom libraries, external APIs, and complex data transformations.

• It lets you build agents that go far beyond the standard constraints of low‑code tools.

• It's ideal for machine learning (ML) workflows, data pipelines, or anything requiring fine‑grained programmatic control.

• You can build your own tools, design your own reasoning loops, and control absolutely everything.

• Essentially: “If you can code it, you can make the agent do it.”

How to Deploy

1. Develop your agent logic using Python in your local environment.

2. Define your application dependencies and package the agent into a container image.

3. Initialize your project using the Azure Developer CLI (azd init).

4. Authenticate your terminal with Azure using az login.

5. Execute azd deploy to automatically build your container, push it to the Azure Container Registry, and provision the necessary cloud infrastructure.

6. Finalize by configuring the dedicated Entra agent identity to manage permissions and access.

Why it matters: Foundry is where you go when you need to break the box entirely, not just think outside it. I’ll get there eventually… maybe one day.

How to Choose the Right Path

Choosing the right tool depends on your team's existing skill sets and the specific needs of your project. If you are a business user looking for quick wins without writing code, Agent Builder provides the lowest barrier to entry. For IT professionals needing to orchestrate multi-step enterprise workflows with scheduling and complex logic, Copilot Studio is the industry standard. However, if your project involves custom AI workloads, advanced data processing, or requires full programmatic control using Python, Foundry is the necessary choice.

Comparing the Paths

To make things a bit easier to digest and see how these tools stack up against one another, here is a quick side-by-side comparison:

Closing thoughts

Best Practices

Regardless of the path you choose, the best way to succeed is to start small. Identify a specific, high-frequency pain point and build a prototype to address it before scaling. Iterate based on user feedback to ensure your agent provides genuine value. Finally, establish clear governance early on to manage permissions, data access, and deployment lifecycles across your organization.

And there you have it—a high-level breakdown of the three distinct paths to building and deploying agents within the Microsoft ecosystem. Whether you just need a quick automation to summarise your morning catch-ups, or a complex Python-driven AI workload, there is a tool designed exactly for that layer of the stack. Don't be afraid to experiment with them, and good luck with your agent-building journey!