Most blocks in Flowrunner follow strict instructions: do X, then Y, then Z. That works well for predictable tasks - but some workflows need flexibility.

That’s where the AI Agent block comes in. Instead of fixed execution, the agent can read your instructions, look at the tools you’ve given it, and decide the best way forward.

Think of it as a smart teammate living inside your flow.

What Makes Agents Different

• Dynamic tool use – the agent decides which actions to run.
  
• Flexible order – it can choose the sequence instead of following a rigid path.
  
• Context awareness – agents remember recent messages to keep track of state.
  
• Adaptive execution – results influence the next step, without manual branching.
  
  

How to Configure

When you drop an AI Agent into a flow, you define:

• AI Provider & Model – OpenAI, Anthropic, Google Gemini, and others.
  
• Prompts – system-level rules and user instructions.
  
• Memory – how much past interaction the agent remembers (default: 15 messages).
  
• Tools – any actions (Slack, Gmail, DB queries, APIs, custom blocks) the agent can call.
  

You decide how much control to keep. Set tool fields yourself for precision, or leave them empty and let the agent fill them dynamically.

Where Agents Shine

• Customer Service – one agent can classify, respond, escalate, or log tickets without needing separate flows.
  
• Approvals – instead of a maze of conditions, one agent can read requests and decide the right path.
  
• Data Processing – agents can analyze, summarize, and pass data forward intelligently.
  
• Workflow Orchestration – combine multiple tools and let the agent decide how to use them in real time.
  

Monitoring and Control

For deeper visibility, Flowrunner integrates with LangSmith, so you can see:

• Which tools the agent picked.
  
• How prompts influenced its choices.
  
• Token usage and performance timing.
  

This gives you insight without losing transparency or control.

The Takeaway

AI Agents give you productivity in a single block. Instead of building long chains of logic, you can give an agent the right tools and instructions — and let it figure out the rest.

👉 If you want your flows to handle complexity without exploding in size, drop in an AI Agent

‍