Test the Workflow Before You Build the Automation

Idea: For a new AI product, test one real customer task manually before building the full agent.

Choose a specific user and a task they already need done. Invite a small number of people to submit one example, explain clearly what will be handled by a person, and return the result within a fixed time window. Ask for permission before using their information, and keep a human responsible for checking the output.

The riskiest assumption is not whether the agent can be built—it’s whether this task matters enough for someone to return to the workflow. Look for repeat use and whether the result genuinely helps; compliments about the concept are weaker evidence.

Set the boundary in advance: one task, one short trial, one review. Continue only if users choose to bring another real task. If they don’t, learn why before automating the wrong thing.