Skip to main content
When you use Clarity to document a process, you get a complete picture of how work happens today and where automation could help. This guide shows how to take that Clarity output and turn it into a working Duvo agent — covering the decisions to make at each stage and an end-to-end example from start to finish.

​
Before You Start

Make sure these are in place:
  • A Clarity process in Complete status with at least one capture analyzed
  • Team settings configured in Settings > Clarity (company name, industry, hourly rate) so the Impact figures are meaningful
  • A rough sense of which part of the process you want to automate first — you do not need a fully scoped plan, but a starting hypothesis helps

​
Step 1: Review the Clarity Output

Open your completed process in Clarity and work through the tabs in this order. Each one gives you something specific you will use when building the agent.

​
Steps

The Steps tab shows a breakdown of each action in the process — who does it, what system they use, what decision they make, and what happens next. Read through the steps and mark the ones that are:
  • Good automation candidates: repetitive actions, data lookups, form submissions, rule-based decisions, and anything the same person does every time the process runs
  • Keep human for now: judgment calls based on context the agent cannot see, sensitive communications, or actions where mistakes are hard to reverse
You are looking for a contiguous sequence of steps that can run without human intervention. That sequence becomes the core of your agent’s AOP.

​
Vision

The Vision tab describes what the process could look like once optimized. This is the end state to aim for, not the first version to build. Read it to understand the goal, then set it aside — you will start with a narrower scope and expand over time.

​
Impact

The Impact tab shows projected financial impact — time saved, cost avoided, and estimated return. Use this to confirm you are automating the right thing first. If the Impact figures are lower than expected, check whether you have captured all the people and systems involved in the process.

​
Risk and Vulnerabilities

This tab lists the exception cases, edge conditions, and failure points identified in the process. Keep this open when you write your AOP — every risk listed here is a case your agent needs to handle, either by resolving it automatically or by pausing for human review.

​
Transformation Guidance

The Transformation Guidance tab contains editable guidance that describes how automation could be applied to this specific process. This is the bridge between the Clarity analysis and your agent. Read through the Transformation Guidance. Edit it to reflect any adjustments you want to make — for example, keeping a step human that the AI marked for automation, or noting a specific system connection to prioritize. The guidance you leave here shapes the context your agent operates in.

​
Step 2: Decide What to Automate First

Start with the smallest slice that delivers meaningful value. A common mistake is trying to automate the full end-to-end process in the first version — this makes testing harder and refinement slower. A good first version:
  • Covers three to seven consecutive steps
  • Has a clear start trigger (a file arrives, a Slack message comes in, a schedule fires)
  • Produces a clear, verifiable output (a row in a spreadsheet, a message in Slack, an email sent)
  • Handles the two or three most common exception cases from the Risk and Vulnerabilities tab
Leave for later:
  • Steps that require significant judgment or context from outside the process
  • Edge cases that happen less than once a month
  • Connections to systems you do not yet have access to

​
Step 3: Build the Agent

​
Open the Agent Builder

  1. Click + Create Agent from your dashboard.
  2. Select Use Agent Builder.

​
Write the AOP

Use the Steps tab, Transformation Guidance, and your own notes to write the AOP. Describe the workflow as you would explain it to a colleague, starting with what triggers the agent and ending with the final output. A practical AOP for a Clarity-derived agent includes:
  1. The trigger — what starts the agent (a schedule, a Slack mention, a file drop, a status change)
  2. The input — what data or context the agent reads at the start
  3. The steps — the sequence of actions from the Steps tab, phrased as instructions
  4. Decision logic — the conditions from the Steps tab that determine which path to take
  5. Exception handling — what to do when the situations in the Risk and Vulnerabilities tab occur
  6. Approval gates — where a human should review before the agent proceeds, based on the high-risk steps you identified
  7. The output — what the agent produces or updates when the run is complete
Click Generate to create the AOP from your description. Review the output, then refine the description and regenerate if it does not match the workflow.

​
Add Human-in-the-Loop Approval Gates

For any step where a mistake would be hard to reverse — sending a customer email, updating a financial record, submitting a form to an external system — add a Human-in-the-Loop step in the AOP: 
Before [action], send a Human-in-the-Loop request showing:
- What you are about to do
- The key data you are acting on
- The expected outcome

Wait for approval before proceeding. If the request is denied, stop and post a summary of what was skipped.
See Designing Human-in-the-Loop Workflows for guidance on where to place approval gates and how to phrase them.

​
Add Connections

Under Connections, enable the connections to the systems involved in the process. Check the Steps tab in Clarity to identify every system that appears — each one needs a connection authorized in the agent. If a system you need is not in the available connections list, see Building Custom Connections.

​
Test

Click Start Work to run the agent against a real or representative example. Watch the live execution view to see exactly what the agent does at each step. Compare what you observe to the Steps tab in Clarity — the agent should follow the same sequence. If the result is not what you expected, return to the Agent Builder, note what diverged, and update the AOP. See Refining Your Agent for a structured approach to iteration.

​
End-to-End Example: Supplier Invoice Follow-Up

The situation: Your procurement team spends about an hour each week chasing suppliers for overdue invoices. The process involves checking a spreadsheet for invoices past their due date, looking up the supplier’s contact email in the ERP, drafting a follow-up email, getting a manager’s approval, and sending the email. Three people are involved but the work is largely the same each time. What Clarity found:
  • The Steps tab showed seven steps, five of which were the same every time regardless of supplier or invoice amount
  • The Risk and Vulnerabilities tab flagged invoices over $50,000 and suppliers with an active dispute as cases that needed extra care
  • The Impact tab projected 3.2 hours saved per week at current team size
  • The Transformation Guidance described an automation that reads the overdue invoice list, drafts follow-up emails using templates keyed to invoice age, and routes anything above $50,000 or flagged as disputed to the procurement manager for review before sending
The scoping decision: Automate the five repeatable steps. Keep the manager review for high-value and disputed invoices, and start with invoices under $50,000 to de-risk the first version. The AOP: 
You are a procurement assistant. Your job is to follow up on overdue supplier invoices.

Run every Monday at 9:00 AM.

Step 1: Read the "Overdue Invoices" tab in the Procurement Tracker spreadsheet.
Collect all rows where the Due Date is more than 7 days ago and the Status is "Unpaid".

Step 2: For each overdue invoice:
  - Look up the supplier's accounts payable contact email in the ERP using the Supplier ID column.
  - If the invoice is over $50,000 or the supplier's "Dispute Flag" column is "Yes", skip it
    and add it to the escalation list.

Step 3: For each remaining invoice, draft a follow-up email:
  - Subject: "Payment reminder — Invoice [Invoice Number]"
  - Body: A polite, professional reminder that includes the invoice number, amount, original
    due date, and a request to confirm the expected payment date.
    Do not threaten or express frustration.

Step 4: Send each drafted email to the supplier's accounts payable contact.
  Update the Status column in the spreadsheet to "Chased - [today's date]" after each email is sent.

Step 5: For any invoices in the escalation list, send a Human-in-the-Loop request to the
procurement manager. Include the invoice number, amount, supplier name, and reason for escalation
(over $50,000 or dispute flag). Wait for the manager to review before sending — do not send
the follow-up email without approval.

Step 6: Post a summary in the #procurement Slack channel: how many invoices were chased,
how many were escalated for review, and the total overdue amount covered.
Connections added:
  • Google Sheets (to read and update the Procurement Tracker)
  • [ERP system via Custom MCP] (to look up supplier contact emails)
  • Gmail (to send follow-up emails)
  • Slack (to post the summary)
Trigger: Weekly schedule, Monday at 9:00 AM First test result: The agent correctly identified four overdue invoices, drafted professional follow-up emails, and escalated one invoice above $50,000 to the procurement manager. One email had the wrong contact email because the ERP lookup returned a primary contact rather than the AP contact — the AOP was refined to specify “accounts payable contact” and the test was rerun successfully. What expanded later: After two weeks of reliable operation, the team added the disputed invoices workflow (a different email template and a different escalation path) and extended the schedule to run twice a week.

​
Tips for Better Results

Use the Steps tab as a checklist: After the first run, compare what the agent did against the Steps tab in Clarity step by step. Any divergence is a specific refinement to make in the SOP. Copy exception handling verbatim from Risk and Vulnerabilities: The exception cases in Clarity are grounded in what your subject matter experts described. Copy the language directly into the AOP rather than paraphrasing — it preserves the nuance. Start the Transformation Guidance with what you want, not what you have: Before copying the guidance into the AOP, edit it in Clarity to reflect your scoping decision. Remove steps you are keeping human, add notes about the connections you will use, and clarify any ambiguous language. The edited guidance becomes part of the AOP context. One scope, one agent: If the Clarity process covers multiple distinct workflows — for example, both routine invoices and disputed invoices — create separate agents for each rather than a single agent with complex branching logic. This makes testing and refinement easier.