LoopAI-103

Go deeper

Unlocks when you've watched the lesson.

0 of 1 watched · about 2 min
Read this lesson

Why a team beats a soloist

One agent has one set of instructions and one prompt, and that only stretches so far. With several agents, each gets its own skill or point of view. You can combine their answers for better decisions, let each step build on the one before, and pass control to whoever fits the moment. Orchestration handles the coordination for you, so you can run analyses side by side, build step-by-step pipelines, or hand work between experts as it goes.

Pipes and stations

Picture an old office building with pneumatic tubes. Capsules shoot between stations, and each station does one job. A workflow in Agent Framework works like that. The stations are executors: the workers that take a message, do something, and send a result on. An executor can be an agent, or plain code. The tubes are edges, which decide where messages go next. A direct edge links one station to the next. A conditional edge only fires when a condition is true, like offering other dates when the hotel is full. A switch-case edge sends each message down one of several routes, like VIP guests to the premium desk. Fan-out sends one message to several stations at once, like checking flights and hotels together. And fan-in gathers their results into one, like a single travel plan. Workflows can also save checkpoints, so a long job can stop and pick up again later.

Lamps on the wall

To see what's happening, workflows send out events, like lamps lighting up on a control panel. Today there's one event type, and you check what it's for: the run started, an executor was invoked or completed, an intermediate result or the final output arrived, a request for input came in, or something failed.

Five ways to coordinate

On top of this, Agent Framework ships five ready-made orchestration patterns. Concurrent sends the same task to several agents at once and collects their answers. Sequential passes the work down a fixed line of agents. Handoff lets agents pass control to each other as the conversation goes. Group chat puts agents in one shared conversation, with an orchestrator choosing who speaks next. And Magentic gives a manager agent a plan to build, delegate, and adapt as it learns. The recipe is always the same: define your agents and what they're good at, pick a pattern and build the workflow with its builder, run it with your task, and read the results. Because every pattern shares that shape, you can try another one without rewriting your agents.

Facts from Microsoft's documentation: Workflow orchestrations in Agent Framework; Agent Framework: edges; Agent Framework: events (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.