What is an Agent?
An Agent in Metaflow is a lightweight, goal-oriented executor that makes decisions, runs Flows, uses tools, and adapts based on context. It’s not just a chat interface. It’s an orchestration layer that knows what needs to be done—and figures out how to do it.
You define its:
- Goal: what it’s responsible for
- Tools: what it can use (Flows, APIs, actions)
- Context: what knowledge or constraints it should consider
Once set up, an Agent acts with autonomy. It can reason, plan, execute, reflect, and repeat.
Why Use an Agent?
Use Agents to:
- Offload recurring tasks like lead qualification, social listening, or copy generation
- Handle decisions dynamically, not hardcoded in a workflow
- Run flows with judgment, choosing when and how to trigger steps
- Wrap complex workflows into simple, reusable digital teammates
- Scale your own thinking by giving each Agent a focused role in your GTM stack
Agents allow you to operate at the goal level—not just the task level.
When to Use an Agent
Use an Agent when:
- The outcome is fixed, but the steps may vary
- You want the system to choose what to do next based on data or reasoning
- You’re orchestrating across tools, steps, or clients
- You’re building reusable, goal-driven automation with logic embedded
Don’t use an Agent when:
- You’re testing a prompt or building a one-off Flow
- You don’t need dynamic logic—just a straight path
How Agents Work in Metaflow
Agents in Metaflow are:
- LLM-powered: They use reasoning (ReAct-style or CoT) to plan and execute
- Tool-aware: They know what tools they have access to, including Flows, scrapers, APIs
- Stateful: They can remember, retrieve, and reuse knowledge across runs
- Composable: They can call Flows, reference Records, and chain together operations
An Agent might:
- Receive a goal like “Summarize our top LinkedIn posts this week”
- Look up what tools it has (e.g. LinkedIn scraper, summarization Flow)
- Run those tools in sequence
- Return a structured summary, auto-saved to a Record
Agents vs Flows
| Agent | Flow | |
|---|---|---|
| Role | Decides what to do | Executes defined steps |
| Nature | Dynamic, context-aware | Fixed, modular |
| Can call | Multiple tools and Flows | Individual tools |
| Best for | Goals, routines, orchestration | Reusable logic, structured steps |
Agents are brains. Flows are muscle.
Key Capabilities
- Set goals and assign tool access
- Chain prompt-based and structured outputs
- Use ReAct-style decision-making
- Handle input/output between Flows and external APIs
- Maintain and retrieve memory from Records
- Can run on schedule, or be triggered from a Flow
How It Fits In
- Agents are autonomous goal-runners
- Flows are callable tools used by Agents
- Canvas is a creative sketchpad agents can populate
- Records store agent output, state, and memory
- Editor lets you craft prompts and schema for Agent input/output
Continue learning
Explore create your first agent, AI marketing agent builder, what is an AI agent, how to build AI agents for adjacent workflows and setup guidance.