Automation
Automation executes a known process with predefined triggers, steps, and outcomes.
Use automation when the path is known, repeatable, and should remain predictable.
Use when
- Repetitive operational tasks
- Approval and routing flows
- Data syncing and reporting
- LLM-assisted extraction inside a controlled process
Avoid when
- Open-ended investigation
- Tasks with unknown next steps
- High-variance research workflows
- Cases where exceptions are the main product value
What automation is good at
Automation is strongest when the steps can be described before the task starts. It turns a repeated process into a reliable system.
An LLM can be part of automation without making the system an agent. The model may classify, extract, summarize, or draft, while the application still controls the path.
When automation is the wrong fit
Automation becomes brittle when every case requires investigation. If the system must choose tools, inspect results, and revise its path, an agentic section may be justified.
Common mistakes
- Calling every LLM workflow an agent.
- Adding autonomy before exhausting predictable automation.
- Letting the model control state transitions without validation.
Next decision
Compare automation with agents. If the steps are known, automate first. If the path is uncertain, define the smallest bounded agentic section.