Agents
In Sim, you build AI agents as workflows that use the Agent block. There's no separate "agent" object to learn — an AI agent is a workflow, and the Agent block is where it thinks.
What you will learn
An agent is a workflow
You build agents by composing workflows around one or more Agent blocks — the same blocks and connections you already know.
The Agent block is the reasoning engine
It is a model call you can shape: customize its messages, give it tools to act, and load skills for guidance.
Composition is the design
Wiring multiple agent calls together is how you compose and design an intelligent process.
An agent is a workflow
This is the one idea to hold onto: an AI agent is a workflow — one built around the Agent block. Everything you learned about workflows applies. You don't leave the model; you add reasoning to it.
Here's the shape of it — a workflow built around an Agent block where it thinks:
Inside the Agent block
The Agent block is a chat with a model, spawned programmatically. Picture a conversation in ChatGPT or Claude: the block's Messages parameter is exactly those messages, except you set them — you decide what instructions, context, and earlier outputs go in. The model reasons over that and returns a result the rest of the workflow can read.
The Agent block is where it thinks — the reasoning engine of your agent. You design the blocks around it to shape how the agent processes data: what it reads before it reasons, what it does with the result after.
Tools and skills
Two parameters extend the Agent block, and both fit the chat-box picture:
- Tools are the functions and blocks the agent can decide to call while answering — search the web, send an email, run another workflow. They're the tool calls the model makes mid-response.
- Skills are prompt modules the model loads on demand — like a manual it looks up only when a task calls for it, instead of stuffing everything into every prompt.
Make a workflow agentic
The clearest way to see an agent isn't to build one from nothing — it's to take a deterministic workflow and let an agent into it. Replace a fixed rule with an agent that decides the path (routing), or augment a step with real reasoning — or both. Same workflow, now with judgment in it.
Composing intelligence
Your ability to compose workflows that make multiple agent calls is central to designing intelligent processes. And every agentic workflow you build becomes a building block — your workspace's library of agents composes into larger AI systems.
Start with one Agent block; the platform scales as your ambition does. The most sophisticated systems are just more of these, composed — more agent blocks swapping in, more reasoning, more reach, all from the same model you learned here.