To quickly test your new task with mock input data, type workflowai task playground [task name]. Optionally, at this point you can add specific instructions about the type of input you want to test with.

This is an opportunity to make sure that the input and output you created still makes sense when used with some (fake) data. The AI will generate a mock input that matches the input criteria defined when you created the task. If you want to proceed with testing with the mock input, you can type y when prompted, to run the test.

To generate a lot of runs in one go, use workflowai task generate [task name]. When prompted enter the number of runs you’d like done at once. Note: in this case, all generated input is automatically accepted.

After testing, if you find the current input and output don’t fit the goals for your task, you can create a new task with an adjusted description. It’s important to note here that this goal here is to check the format of the input/output, not the content itself.

Once this is done, you have completed one task run!

Example scenario (cont’d):

You want to make sure to test this task with a flight event, since the arrival time for a flight is really early compared to the arrival time of a lot of other types of events.

In order to create mock input data like this, you could enter a description like this: Generate a calendar event that is an airplane flight.

The mock input that the AI generates for you is:

  • Title: Flight from SFO to IAD
  • Description: Delta Flight from SFO to IAD at 9am
  • Location: San Francisco International Airport

The output that is generated is arrival_time_in_minutes = 30

While the output itself does not match your expectation, the format (time in minutes) does. In this case, you should not try to create a new task. Instead, you should annotate this run (see below) to inform the AI that the number of minutes should be different.