Deterministic vs Probabilistic Steps

In automations, steps behave in two main ways: deterministic or probabilistic.

Deterministic steps

Same inputs → same outputs. Re-running the step with the same data produces the same result.

Examples:

• Triggers — Manual, Scheduled, and most event triggers (e.g. new email, new row) return structured data that doesn’t change for that run.
• Data steps — Get rows from a sheet, retrieve emails, scrape a URL: the data is fixed for that execution.
• Utilities — Create list, constant, strip HTML, Freemarker template: output is fully determined by inputs.
• Integration actions — Create/update a record, send email, create calendar event: the action is repeatable with the same inputs.

Use deterministic steps when you need reproducibility, testing, or branching on exact values.

Probabilistic steps

Same inputs → possibly different outputs. The step uses an AI model, so multiple runs can produce different text or choices.

Examples:

• Generate Text — Same prompt can yield different wording each time.
• Extract Information / Summarize / Classify — AI output can vary slightly.
• Assistant / Assistant Structured Data — The Assistant’s reply can differ between runs.

Use probabilistic steps when you need creativity, interpretation, or natural language. For consistency, you can lower temperature or use more specific prompts; for strict reproducibility, prefer deterministic steps or fixed templates where possible.

Why it matters

• Testing — Deterministic flows are easier to test; probabilistic steps may need multiple runs or snapshot comparisons.
• Variables — Both types produce outputs you can reference in later steps (e.g. #StepName.field#). See Variables.
• Conditional paths — Branching on AI output is fine, but remember the value can change on re-run. See Triggers, Actions, and Conditional Paths.