# Agentic workflow

An agentic workflow is a multi-step process in which an AI agent plans the steps, calls tools to carry them out, checks each result and decides what to do next, instead of running every case through the same fixed script.

## How is an agentic workflow different from traditional automation?

Traditional automation follows a path someone drew in advance: when X happens, do Y, then Z. It is predictable and cheap, and it breaks the moment the input doesn't match the path.

An agentic workflow starts from a goal instead of a path. The agent decides which steps to take, in what order, based on what it finds along the way.

| | Rule-based automation | Agentic workflow |
|---|---|---|
| Starts from | A trigger and fixed steps | A goal in plain language |
| Handles surprises | Fails or routes to a person | Adapts the next step |
| Best for | High-volume, identical cases | Varied cases that need judgment |
| Cost per run | Low and fixed | Higher and variable |

Most real systems mix the two: fixed steps where the path is known, agent decisions where it isn't.

## How does an agentic workflow run, step by step?

Take a customer success manager in a customer success platform who asks: "Find every enterprise account whose admin hasn't logged in this month and draft a check-in for each account owner."

1. **Plan.** The agent breaks the request into steps: find enterprise accounts, check admin activity, look up owners, draft messages.
2. **Act.** It calls the accounts API with a plan filter, then the activity API for each admin.
3. **Check.** Two accounts have no admin on record. The agent flags them separately instead of guessing.
4. **Adapt.** It reads recent tickets for the remaining accounts so each draft mentions something specific.
5. **Hand back.** It shows the drafts for review and sends nothing until the manager approves.

A script could run the lookups in step 2 if someone wrote the plan in advance. Steps 3 and 4 are where the agent earns its place: it notices a gap and changes course.

## When should a workflow be agentic?

Use an agent when the inputs vary, the right next step depends on what the last one returned, and a person would otherwise do the job by clicking through several screens. Keep it rule-based when every case looks the same.

Common design patterns help keep these workflows predictable: chain steps where the order is known, route requests to specialized handlers, let one agent split work across sub-agents, and have a second pass check the output. Add a person at any step that spends money, deletes data or contacts a customer.

Inside a software product, the useful version is simple: the user states the outcome and an in-product agent runs the steps through the product's own actions. Foldspace's Product Agent works that way in the product itself, as the signed-in user and with that user's access, and only for [the actions and knowledge a team has defined](https://docs.foldspace.ai/privacy/llm-governance/).

## Learn more

- [Product Agent](/platform/product-agent)