If your AI program is still teaching people to write prompts, what should it be teaching instead?

Welcome to Episode 11 of The Business AI Playbook, the podcast where I sit down with leaders who are working out how AI changes the way their people learn and work.

Listen to the full episode on Spotify, YouTube, or Apple Podcasts.

Prompt training had a shelf life of about two years

Kevin Bishop leads talent development for LinkedIn’s corporate functions: talent, finance, legal and public policy, marketing, and communications. Last fiscal year he also ran enterprise AI enablement. He got there by an unusual route. He trained as a singer and actor, moved to Chicago to perform, and “accidentally got a job at Deloitte” to pay the rent. Years of sales operations spreadsheets pushed him into coaching and then L&D. The views in this conversation are Kevin’s own, not LinkedIn’s official position.

His first point is one I don’t hear L&D leaders say plainly enough:

“We’re well past teaching prompt writing at this point.”

Two or three years ago, prompt training made sense. The models were weaker and people were nervous, so teaching someone how to phrase a request got them a better answer. Today you can ask the AI to write the prompt for you. There’s little point building a program around something the model now does for you.

What I liked about Kevin’s answer is that he didn’t stop at “prompts are done.” He walked through what came next on his team, and what he thinks comes after that.

In this episode, we discussed

  • Why prompt-writing training has run its course, and what replaced it on Kevin’s team
  • Mission AI: seven ways to earn a badge, and why four badges made you an “AI pioneer”
  • Why anxiety was the first blocker to AI adoption
  • The AI tool that took longer to build than two years of doing the task by hand
  • Starting with the problem, not the symptom or the solution someone already picked
  • Why AI enablement is moving into function-specific work
  • How managers lead their teams through changes they can’t fully predict yet
  • Taste, judgment and discernment, and how people build them when AI does the entry-level work
  • Kevin’s route from theater to Deloitte to L&D

Phase two: give people permission to slow down

Once the basics were in place, Kevin’s team found the real blocker was anxiety. People were told to use AI to be more productive, with no direction on what that meant, and without slowing down the work they were already doing. In Kevin’s words, “we didn’t necessarily give people permission to slow down and play a little bit with AI to figure out how it was going to benefit their work.”

So his team built a program called Mission AI. It offered seven different ways to earn a badge, including hackathons, self-paced learning and live sessions, and many of them didn’t feel like training at all. Earn four of the seven and you became an “AI pioneer,” with recognition from executive leadership, points you could cash in for prizes, and a place at a virtual AI summit with outside speakers that ran over a couple of days.

The design choice I’d steal is the seven routes. One mandatory course tells everyone the same thing at the same pace. Seven optional routes let a skeptic start with a hackathon and an enthusiast go straight to building, and both end up in the same place. Kevin’s team got the result they were after: people told them their anxiety was down and they finally had somewhere to start.

The turn: what the 30-minute tool taught them

Then the message changed. A year ago the instruction was to go use AI everywhere. Now, Kevin says, it’s to use AI judiciously, because there is a financial cost and an environmental cost to all of it.

He told a story an executive shared with his team. An employee proudly showed off an AI tool they had built. It took a task that used to take 30 minutes every month and did it in three. The executive asked one question: how long did it take you to build it? The answer was longer than two years of doing the task the old way.

Do the math and the problem is obvious. Saving 27 minutes a month is about five and a half hours a year. A tool that took even a few weeks of work to build needs years to pay back, assuming the task doesn’t change in the meantime.

The part that should worry L&D leaders is that people are bad at judging this about themselves. In a 2025 randomized trial by METR, 16 experienced open-source developers took 19% longer on real tasks when they used AI tools, and afterwards believed AI had made them 20% faster. A follow-up study METR ran in 2026 was inconclusive. The gap between how fast people felt and how fast they actually were is the finding I’d keep. If your people can’t tell whether AI is saving them time, “use AI more” is a weak instruction.

This is why Kevin starts every AI conversation with the business the same way he starts every other one: what problem are you actually trying to solve? Most people answer with a symptom, or with a solution they’ve already picked. Dig in and sometimes AI is the answer. Sometimes the honest answer is that a different workflow, or no tool at all, is faster.

See it in action

Interactive Soft Skills Training - created with Colossyan

Try it yourself →

Phase three: enablement moves into the work

This is where Kevin thinks AI enablement goes next, and I agree with him. Fewer general AI trainings and fewer sessions on tools. Instead, his team sits down with each function’s leaders and asks for the three to five things in their AI transformation plan for the year, then builds learning around how that specific work changes. As Kevin put it, “the AI piece is embedded in the skill development for the larger workflow.”

The programs blend the tool with the human skills around it. If a finance team adopts a new tool, the learning covers the tool, and also the courage it took to question how the work was done and the creativity it took to redesign it. The tools will keep changing, so what his team is really teaching is how to judge whether a new tool improves your workflow.

One honest admission from Kevin stuck with me. Managers can’t reassure their teams with a plan, because nobody has the plan yet. His manager can’t tell him exactly how AI will change his job. In his words, “that’s on the two of us to figure out together.” I think more leaders should say that out loud instead of pretending to have the answer.

The question L&D hasn’t answered yet: where does taste come from?

The most useful part of the conversation was about taste. Kevin’s definition is the simplest I’ve heard:

“Taste is knowing what good looks like.”

He judges an AI-drafted program design against what has worked for him before. That judgment came from years of doing the entry-level work himself. Now AI is taking on exactly that entry-level work. So how does the next person build the judgment to know whether the AI’s output is any good?

Getting this wrong already has a measurable cost. A BetterUp Labs and Stanford Social Media Lab survey of 1,150 US desk workers, published in Harvard Business Review, found that 40% had received AI-generated “workslop” in the past month: output that looks like work but doesn’t move the task forward. Each one took nearly two hours to deal with. About half of the people who received it saw the sender as less capable, and 42% saw them as less trustworthy. People who can’t tell good AI output from bad are already paying for it with their reputation.

Kevin’s early answer is that L&D will lean much harder on scenarios: deliberate practice that replaces the reps people no longer get in their day-to-day work. I think that’s right, and I don’t think anyone has fully solved it yet. It’s also how we think about building at Colossyan: AI does the first draft, and a person with taste decides whether it’s good.

If you run L&D or talent development, I’d like to hear how you’re answering that question. Where will your junior people learn what good looks like?

Case Study

How Paramount cut training video production from weeks to hours

Paramount replaced manual video production with Colossyan's AI avatars, helping its global L&D team create training content faster without traditional production bottlenecks.

Read the full story →

The full episode is now available on Spotify, Apple Podcasts and YouTube.

About the guest

Kevin Bishop is Director of Talent Development at LinkedIn, supporting the company’s corporate functions: talent, finance, legal and public policy, marketing, and communications. Last fiscal year he also led enterprise AI enablement, including the Mission AI program. Before that, Kevin worked in business development and sales operations at Deloitte, then moved into coaching and learning and development. He trained in voice performance and theater, and says that background shapes how he approaches the human side of human + AI. Connect with him on LinkedIn: linkedin.com/in/kevinrbishop.