Test run an agent before you turn it on

Google Workspace Flows productivity guide

Before you turn on an agent to run automatically, you can run it once using real data to check the results. This helps you see if it does what you want. You can also use a test run to run an agent manually if you don’t want it to run automatically.

Important: A test run does real actions. For example, it sends messages, sets up calendar events, and changes docs. If you’re worried about bothering other people or changing important files, see the tips on this page for how to do safer test runs.

Do a test run

  1. Go to https://flows.workspace.google.com and create an agent.
  2. (Optional) Change your agent a little so it doesn't affect other people or real files. For tips, on this page see Tips for a safer test run.
  3. When you’re ready, click Test run.
  4. If your agent starts with On a schedule, click Start.
  5. If your agent starts in other ways, it needs existing data to use.
    1. Sometimes, a default value is automatically filled in for you. For example, with When I get an email, it finds the last matching email you got. If that works, click Start. If you want to start from something else, remove what’s there, click the box, and select the one you want. Then click Start.
    2. Sometimes, you need to select the data yourself. For example, with When I’m mentioned in spaces set to all spaces, you select a space, then a matching message is found automatically. Then click Start.

Review test run results

You can see the test run results as it goes. Some results, like a response from Gemini, are shown right there. Others, like a message in Chat, have a linked chip that you can click to see it in Chat.

You might see the following error cases:

  • Red "Test complete": The test run went as far as it could, but couldn’t finish because something’s wrong with the agent’s setup. Check for issues like missing required fields, incomplete conditional statements, and unconnected integrations, then try again.
  • Yellow "Test couldn’t complete": The test went as far as it could, but couldn’t finish because of an issue with the data it’s trying to use. For example, a security risk was blocked while running a Gemini step. Your agent might be set up correctly, but the test couldn’t run all the way through. Try using different starting data, like a different email to summarize, and try again.

Tips for a safer test run

You might not want your test run to affect other people or real files. Here are some ways to set up your agent to make that less likely. After it works in the safer setup, you can add people or real files before you turn it on.

  • Have emails and chat messages sent only to you, instead of using a variable that could include other people.
  • Use test documents, like a copy of the document or sheet you want your agent to update.
  • Create a test meeting where you’re the only guest.

Was this helpful?

How can we improve it?