When Does AI Make a Task Easier?

Victor Yocco asks teams to look closely before adding AI. A tool may write something faster than a person can. But someone may still need to read it and check it. This can take as long as doing the work from the start. The team should count that time when it compares the two ways.

Yocco gives an example from a company that provides health insurance. Its first tool wrote full letters about people's requests for another review. Staff had to check the letters before they could sign them. Some stopped using the tool, and others took longer on each case. The team changed the tool to find useful documents instead. Staff could then spend less time searching and write the letters themselves.

Yocco suggests four checks before the team starts to build. First, find out how much work the tool adds or removes. Second, compare it with the way people already do the job. Third, ask whether users can see mistakes and what mistakes would cost. A tool can help by showing where it found each piece of information.

The last check looks at what happens after people use AI regularly. Can they still tell whether its answer is good a year later? If needed, let people work through some cases without AI help. This gives them time to use the skills they need for checking. Teams can then use the four answers to choose what AI should do.