# Build a workflow of fixed stages

Build a tender response workflow whose four stages run in the order you set, with a named instruction for each and an approval before the response is released.

In a pipeline you fix the order in advance: each stage does its part and passes its output to the next. On Katonic it is a workflow, an agent built as steps that run in a fixed order. This tutorial builds a tender response: extract the requirements, find past answers, draft, check, and have a person approve before the response is released. It is the seventh form in [Which builder for which agent](/agents/choose-a-builder).

**Who can use it:** the **Builder** role or higher, in a tenant licensed for agents and knowledge.

## Prerequisites

- A knowledge source that holds past bids; see [Add a knowledge source](/knowledge/knowledge-bases/add-a-knowledge-source).
- A team you belong to, for the **Owner team**. [Before you build a form of agent](/agents/forms/before-you-build) says who provides it.

## Build the workflow

  
**Step 1. Start a blank workflow**

In Studio, open **My Agents** and select **New agent**. Select **Automate a process**, pick the **Blank** card and select **Continue to build**.

  
**Step 2. Name the flow and pick the owner**

On **Identity**, type the **Workflow name** `Tender Response` and a **Description**, and pick the **Owner team**.

  
**Step 3. Keep the trigger manual**

Open **Trigger** and leave the **Manual** card selected, so that a person starts a run with the tender's text.

  
**Step 4. Add a step for each stage**

Open **Steps**. A blank flow starts with one **LLM** step, a step that sends text to a model. Open it and fill in the first block below. Then, under **Add a step**, select **Knowledge**, **LLM** and **LLM** in turn, filling in each as you add it. Steps run in the order listed, and a new step is added after the last one.

    ```text
    Step name: Extract the requirements
    Instruction: List every requirement in the tender below, one per line, numbered. Copy each requirement in the tender's own words. Do not answer them.

    Step name: Find past answers
    Knowledge base: the source that holds past bids

    Step name: Draft the answers
    Instruction: For each numbered requirement, write an answer from the past answers you are given. Where no past answer covers a requirement, write NO PAST ANSWER and nothing else.

    Step name: Check the draft
    Instruction: Compare the draft with the numbered requirements. List each requirement that has no answer, or an answer that does not address it. Change nothing in the draft.
    ```

    You should now see four named steps.
    ![The workflow builder on Steps for the Tender Response draft: four named steps in order, Extract the requirements, Find past answers, Draft the answers and Check the draft, the first one open with its Step name and Instruction filled in, and the Add a step row below.](https://docs.katonic.ai/img/agents/forms/fix-the-order-of-the-work/forms-pipeline-steps.png)
*Steps holds the work in the order it runs. Name each step for the stage it does and give each LLM step its instruction.*

  
**Step 5. Add the approval**

Open **Branches & approvals** and select **Add approval**. In **Approval label**, type `Bid manager sign-off`. In **Reviewers / channel**, type the channel to notify, here `#bids`. The run pauses there until a reviewer decides.

  
**Step 6. Check the flow on the canvas**

Select **Visual** in the build header. The canvas shows the same draft as nodes, one box for the trigger, each step, the approval and the output, joined in the order they run. Select a node to open its configuration. On an **LLM** node, **System prompt** holds the instruction you typed, **Insert variable** lists the upstream values it can refer to, such as the trigger's message, and **Pass the previous step's output in** hands the step what the one before it produced.
    ![The Visual canvas for the Tender Response draft: a trigger node, the four named steps and an approval node joined in a line to the output node, the Draft the answers node selected, and the inspector on the right open on that node's Config tab.](https://docs.katonic.ai/img/agents/forms/fix-the-order-of-the-work/forms-pipeline-canvas.png)
*The Visual canvas shows the same flow as nodes joined in the order they run. Select a node to open its configuration.*

  
**Step 7. Name the output and set the guardrails**

Select **Wizard**. Open **Output**; in **Output label**, type `Tender response`, and in **Destination**, `return to caller`. Open **Guardrails** and switch on **Cost ceiling per run**.

## Test and publish

  
**Step 1. Run the draft**

Open **Test**, paste three or four requirements from a tender as the message and select **Run**. Every step reports in the **Run trace** with its status, timing and output. The sandbox continues the approval automatically.

  
**Step 2. Publish**

Select **Publish** and fix any blocker. The screen that follows says whether the version is live, in a lane or not live yet; see [Before you build a form of agent](/agents/forms/before-you-build).

  
**Step 3. Run it for real**

Open the workflow from **My Agents**, select **Run**, paste the same requirements into **Input** and select **Run now**. Members start a run the same way from the Work Room; see [Start an automation](/user-guide/work-room/start-an-automation).

## Verify

Use a tender with three requirements, one of which no past bid covers.

- In the Test tab, the trace lists the four steps in the order you set. The first step's output is three numbered requirements. The third step's output has an answer for two of them and NO PAST ANSWER for the third. The fourth lists the third requirement as unanswered.
- On the **Run** tab after publishing, the run stops at **Bid manager sign-off** and reads as awaiting approval. The person who started the run cannot approve it; a reviewer does, from **Automations** in the Work Room.
- After the approval the run completes, and its output is saved to Artifacts.

## Troubleshooting

- **A step produced nothing and the run still completed.** A later step can only work from what it was given. Open the run, find the first step whose output is empty or wrong, and fix that step's instruction or its source.
- **A stage does not see the earlier output.** On the canvas, select the node and check that **Pass the previous step's output in** is on, or use **Insert variable** in its **System prompt** to refer to the upstream value.
- **The run stays at the approval.** Nobody has decided. See [Automations and approvals](/user-guide/work-room/automations-and-approvals).
- **The flow cannot be published.** A **Visual** node on the canvas, the chart node marked coming soon, blocks publishing until you take it off. So does a step that writes with no approval step before it.
- **A step is in the wrong place.** A new step always goes after the last one. Remove the steps after the right place with **Remove step** and add them again in order.
- **A coordinator cannot route to this workflow.** A published workflow cannot take a delegated turn. Run it directly.

## Related

- [Which builder for which agent](/agents/choose-a-builder)
- [Automate a process](/agents/automate-a-process)
- [Automations and approvals](/user-guide/work-room/automations-and-approvals)
- Next form: [Call other agents from one agent](/agents/forms/call-other-agents-from-one-agent)

Source: https://docs.katonic.ai/agents/forms/fix-the-order-of-the-work
