In brief
Ontologie offers two automation engines: Workflows and Agents. This guide helps you choose which one to use based on your needs.
What you will learn
- The fundamental differences between an agent and a workflow
- When to favor one over the other
- How to combine them for advanced use cases
Quick comparison
| Criterion | Workflow | Agent |
|---|---|---|
| Logic | Deterministic: steps are defined in advance | Autonomous: the agent reasons and decides at each step |
| Trigger | Automatic (cron, webhook, event) or manual | Conversation or API call |
| Inputs | Structured (variables, triggers) | Natural language (questions, instructions) |
| Outputs | Predictable and reproducible | Variable depending on context and reasoning |
| Error handling | Managed by dedicated blocks (retry, error_handler) | The agent adapts its strategy |
| Complexity | Processes with clear rules | Vague or exploratory tasks |
| Monitoring | Per execution (success/failure/logs) | Per session (reasoning, tools used) |
When to use a Workflow
A workflow is the right choice when:
- The process is repetitive : The same steps repeat with each execution.
- The rules are clear : You know exactly what to do in each case (conditions, loops, filters).
- Reliability comes first : You need identical results with each execution.
- The trigger is automatic : An event, a schedule or a webhook starts the process.
Examples:
- Send a welcome email when a customer is created
- Synchronize data between two systems every hour
- Calculate a risk score from defined business rules
- Generate a weekly report and send it by email
When to use an Agent
An agent is the right choice when:
- The task is exploratory : You don't know in advance how many steps will be needed.
- The input is in natural language : The user asks a question or gives a free-form instruction.
- Reasoning is required : The agent must analyze, compare, and synthesize information.
- The context changes : The answer depends on the ontology, documents, and conversation history.
Examples:
- "Which suppliers deliver to Europe with a lead time under 5 days?"
- "Summarize the penalty clauses in our contracts with ACME Corp"
- "Compare the performance of our three warehouses this quarter"
- "Suggest a sourcing strategy for the new product X"
Combining them
In advanced cases, you can use both together:
- Workflow → Agent: an
ai_agentblock in a workflow delegates an analysis task to an agent. - Agent → Workflow: an agent uses the
workflowtool to trigger a structured process.
Example: a workflow detects an anomaly in the data (deterministic logic), then calls an agent to analyze the cause and write a report (reasoning).
Expected outcome
You know when to use a workflow (repetitive processes, clear rules, automatic triggers) and when to use an agent (open questions, contextual analysis, exploration). You can also combine them for complex scenarios.
Next steps
Need help?
Contact us: Support and contact.
