AX-200 · Lesson 9 of 10 · Module 3: Run It Well · 3:08
When it breaks
Workflows break, often silently. Send every error to a named person and a deputy, keep a fallback that hands the job back to people, and look at the run history every week. Give the workflow only the access its steps need, keep customer data in approved tools, and treat what customers write as information, never as orders: a message can contain text that tries to steer the AI, so nothing costly should happen without a person. Worked example: a renamed form field stopped Tidewell’s workflow for two days while the error emails went to an inbox nobody reads.

What you will learn
- Set an alert and a fallback for a workflow
- Give a workflow only the access it needs
- Treat outside text as information, not orders
The one thing to remember
Make failures loud, and keep a way back.
Common mistake: Assuming no news means it’s working
Workflows can stop or misfire quietly, so the first sign is often a customer who has waited for days.
The fix: Send errors to a named person, keep a fallback, and check the run history every week.
Key terms
- Error alert
- A message to a named person when a run fails.
- Fallback
- How the job still gets done when the workflow stops.
- Least access
- Giving a workflow only the access its steps need.
Part of AI Workflow Specialist (AX-200)
For teams with repeat work: find the jobs worth automating, build the workflow with no code, then test and measure it.
10 lessons, practice quizzes, a final exam and the Sophyx Certified AI Workflow Specialist – Associate certificate. Regular price USD 49.99. Your first certification is free. Free with any paid Sophyx plan.
Start the course free