AI Agent Examples

Examples of how AI agents can be used in real-world scenarios, including task automation, decision-making support, and multi-step workflows.

Latenode's AI Agent is flexible enough to support multiple architectures: from a single dynamic assistant to modular multi-agent systems and external knowledge integration.

1. AI Agent Basic Workflow Example

A single AI Agent receives user prompts and decides which tools to use, if any. This setup is lightweight but powerful — capable of parsing, routing, and composing responses dynamically.

Scenario Structure

One central AI Agent

Connected to:

weather_tool (e.g. wttr.in)

exchangerate_tool (e.g. exchangerate.host)

web_search_tool (e.g. factual search)

Input from Trigger, output to SetVariables

Call Example 1 — Weather + Currency

Prompt:

“What’s the weather in Berlin and how much is 100 EUR in USD?”

The agent triggers:

weather_tool with the city "Berlin"

exchangerate_tool for EUR to USD

Skips unrelated tools

Expected output:

It's currently 17°C in Berlin. 100 EUR is about 108 USD.

Call Example 2 — Simple Fact

Prompt:

“Who is the CEO of Apple?”

Agent skips weather and currency

Only triggers web_search_tool

Expected output:

The CEO of Apple is Tim Cook.

🎯 This scenario is great for lightweight assistants that respond contextually without complex logic trees.

🔗 You can copy this template here: AI Agent Basic Workflow Example

2. Multiagent - AI Multi-Agent Interaction Example

This approach uses a main agent to break down user requests and forward sub-tasks to specialized agents. Each sub-agent operates independently and can have its own API logic.

Scenario Structure

main_agent controls the overall logic

Delegates to:

weather_agent

finance_agent

web_search_tool

Each sub-agent is connected to dedicated APIs or logic blocks

Call Example 1 — Weather + BTC

Prompt:

“What’s the weather in Tokyo and what’s the BTC price?”

main_agent sends:

Weather part → weather_agent

Bitcoin price part → finance_agent

Expected output:

It's 24°C in Tokyo right now. BTC is trading at $68,200.

Call Example 2 — CEO + HQ Weather

Prompt:

“Who is the CEO of Apple and what’s the weather like at their HQ?”

Agent parses:

Apple HQ location → via web_search_tool

Weather in that location → via weather_agent

Expected output:

The CEO of Apple is Tim Cook. It’s 21°C and sunny in Cupertino.

🧩 This pattern fits well for scalable assistants, where logic needs to be cleanly split.

🔗 You can copy this template here: AI Multi-Agent Interaction Example

3. AI Agent with Cloudflare AutoRAG Database

Integrate an AI Agent with Cloudflare AutoRAG to retrieve structured external knowledge — such as product documentation, policies, or internal data.

Scenario Structure

cloudflare_rag_agent handles free-form prompts

Two HTTP tools connected:

rag_database_docs — deep semantic retrieval

raw_data — fast, factual lookups

⚠️ Before using this scenario, you must:

✅ Most modern RAG platforms - including AutoRAG - automatically generate embeddings server-side. You don’t need to preprocess documents or manage vectors manually.

Call Example 1 — Documentation Question

Prompt:

“How does the billing system of Cloudflare work?”

Agent detects it’s a high-level question

Selects rag_database_docs to retrieve semantic context

Responds based on indexed content

Expected output:

Cloudflare's billing system uses a monthly subscription model with pro-rated charges...

Call Example 2 — Quick Data Point

Prompt:

“What’s the max bandwidth on the free Cloudflare plan?”

The agent determines it's a factual request

Selects raw_data for direct value retrieval

Expected output:

The free Cloudflare plan includes up to 1 TB of monthly bandwidth.

📘 Use AutoRAG-style integrations for assistants that can reason over your documents and give context-aware, accurate replies — without hosting your own vector database or embeddings pipeline.

🔗 You can copy this template here: AI Agent with Cloudflare AutoRAG Database

Best Practices

Name all nodes descriptively — they become visible "tools" to the agent

Use Agent ID to maintain session-based memory

Set Max Iterations to prevent loops

Use Output JSON Schema if the response needs to be structured

Did this answer your question?

😞

😐

🤩