Moving AI Beyond the Pilot: Part 2 — The Employee Side
There is a micro-side that an individual employee or manager can do to ensure the successful AI deployment will deliver on its promises.
Last night, I went mountain biking and left my fifth-grade son at home to finish his math homework. He’s a great problem-solver who enjoys complex questions. Practicing basic fraction arithmetic? Apparently beneath his pay grade.
When I got back, he proudly announced he was finished and asked to play Xbox. Off he went, while I checked his homework. Of the 20 fraction calculations, three were wrong, with no work shown. On the five more complex word problems, he’d shown some work but also gotten three wrong.
I called him back and asked him to redo the three basic calculations. This time, he showed his work—and still got all three wrong. How had he managed the other 17 in his head?
After some discussion, the truth emerged: Alexa had done the arithmetic. His defense? “I already know how to do that. Why should I waste time practicing?”
The next hour we spent working through fractions together, suggested otherwise. His shortcut had become a father-son workshop. Neither of us had budgeted time for it.
So what does this have to do with getting AI beyond the pilot stage at a biotech or pharma company? Quite a bit.
1. A confident answer still needs a reality check. One answer suggested that 1/2 + 3/4 = 4 1/4. Whatever happened between his question and Alexa’s response, a quick estimate should have raised a flag. At work, employees need enough understanding—and permission to pause—to question an answer that doesn’t make sense.
2. Check what went in. Alexa may have misheard the question. We couldn’t establish exactly what happened, but the distinction matters: a tool can answer the wrong question perfectly. Missing context, ambiguous instructions, or incorrect source data can create the same problem at work.
3. Don’t outsource the understanding you need to review the result. He could identify the steps in the harder problems, but his shaky arithmetic let him down. Employees also need enough subject knowledge to evaluate AI’s work. Saving time on a task is useful; losing the ability to recognize a mistake creates more work later.
4. Be clear about the actual goal. This was the biggest lesson. His goal was to finish the worksheet and unlock Xbox. My goal was for him to learn fractions. We had a slight alignment problem.
At work, “finish this faster” can create the same disconnect. Managers need to explain what a good outcome looks like, what needs checking, and where human judgment matters. Employees need to understand that producing an answer and taking responsibility for it go together.
In Part 1, I discussed leadership, technology, and culture. This is where those ideas meet everyday behavior: what someone asks AI to do, what they check, and whether the result actually serves the purpose.
My son did learn something that evening. So did I: before celebrating productivity gains, check the fractions.
When I got back, he proudly announced he was finished and asked to play Xbox. Off he went, while I checked his homework. Of the 20 fraction calculations, three were wrong, with no work shown. On the five more complex word problems, he’d shown some work but also gotten three wrong.
I called him back and asked him to redo the three basic calculations. This time, he showed his work—and still got all three wrong. How had he managed the other 17 in his head?
After some discussion, the truth emerged: Alexa had done the arithmetic. His defense? “I already know how to do that. Why should I waste time practicing?”
The next hour we spent working through fractions together, suggested otherwise. His shortcut had become a father-son workshop. Neither of us had budgeted time for it.
So what does this have to do with getting AI beyond the pilot stage at a biotech or pharma company? Quite a bit.
1. A confident answer still needs a reality check. One answer suggested that 1/2 + 3/4 = 4 1/4. Whatever happened between his question and Alexa’s response, a quick estimate should have raised a flag. At work, employees need enough understanding—and permission to pause—to question an answer that doesn’t make sense.
2. Check what went in. Alexa may have misheard the question. We couldn’t establish exactly what happened, but the distinction matters: a tool can answer the wrong question perfectly. Missing context, ambiguous instructions, or incorrect source data can create the same problem at work.
3. Don’t outsource the understanding you need to review the result. He could identify the steps in the harder problems, but his shaky arithmetic let him down. Employees also need enough subject knowledge to evaluate AI’s work. Saving time on a task is useful; losing the ability to recognize a mistake creates more work later.
4. Be clear about the actual goal. This was the biggest lesson. His goal was to finish the worksheet and unlock Xbox. My goal was for him to learn fractions. We had a slight alignment problem.
At work, “finish this faster” can create the same disconnect. Managers need to explain what a good outcome looks like, what needs checking, and where human judgment matters. Employees need to understand that producing an answer and taking responsibility for it go together.
In Part 1, I discussed leadership, technology, and culture. This is where those ideas meet everyday behavior: what someone asks AI to do, what they check, and whether the result actually serves the purpose.
My son did learn something that evening. So did I: before celebrating productivity gains, check the fractions.
Browse by tag
All
AI Clinical Trials FDA IND Submission IT Inspection Readiness ai ai deployment biotech business-transformation cio commercialization computer vision corporate-strategy culture digital-transformation entrepreneur entrepreneurship high-performing team leadership life-sciences outcome-focus pharma psychologically safe regulatory risk software-engineering sourcing sports transformation