Primary content and writing by Steve Nackers, Program Director, UW–Madison, Interdisciplinary Professional Programs – Draft assistance and review via Claude and Gemini
When I evaluate training, books, or speakers for AI content, I ask myself one question. If every featured tool is gone next month, what can I still do?
Not much of what I see organizations and learners buy survives that question.
When the Training Misses the Room
Sit in on enough AI rollouts and vendor calls and a pattern starts to emerge. You see a lot of depth or specificity, transformer architecture, embeddings, context windows, new agents or interfaces. A large group of mixed roles are present and trying to absorb what is put in front of them.
Each element can be valuable, and in fact, some people need all of it. However, in most of the sessions I’ve attended, there are only a handful of people who are going to design a model. The rest are trying to find ways to apply the technology safely and effectively for their needs, industry, or role.
Park, Skid, or Track
People are looking for some form of “Driver’s Ed,” and not everyone is at the same level. Some need to learn to parallel park. Baseline competence. What the tool is for, how to specify what you want, and when not to use it at all.
Some people need to know how to handle a skid. Their work has consequences when AI use goes wrong, and they need to know how to course correct.
A few want to track instruction. They’ll build the workflows other people end up using and search for the technological edges on purpose.
Then there are the people who build what’s being driven. That’s a profession of its own with real depth. It isn’t the top of a single ladder everyone else is climbing, and treating it that way is like teaching engine mechanics to drivers or transformer model theory to someone looking to summarize their week’s emails effectively.
The Group That Isn’t Driving
There are individuals who don’t spend the majority of their time in the driver’s seat. They are directing the driver, or responsible for supporting the team. In business or education, these could be managers, reviewers, technical leads, administrative staff, and more; people who don’t produce the work but still share responsibility.
Being able to validate and review work has always been a primary challenge for leadership. This has become even more difficult with the fluency of AI. You own the output your team produces with AI just as you own the output it produces without AI. Which means the questions are yours to ask. “Where did this come from? What was checked? How much did it cost? What happens if we’re wrong?” We need to equip the people directing and reviewing work to ask them.
Reviewing takes as much skill as doing, and we’ve trained for it for decades through audit, quality systems, and design review. The difference now is that a wrong answer reads as well as a right one.
What Driver’s Ed Actually Teaches
When people ask me how to get better at this, I reach for my kids.
“Go clean your room,” leads to wildly varying results. “Pick up the clothes, put dirty ones in the laundry and clean ones in the dresser, then dust, vacuum, and throw out the trash,” gets a better, more consistent result. Same kid, same room, but clearer directions.
For a given model, on a task it is capable of, changing how I specify the work has moved my results more than any other change. You learn to check your mirrors once and it comes with you into every car you ever drive. Same with being able to say what you want and what finished looks like.
The other half of managing AI work is reading the road. In a research program from 2023, Dell’Acqua, Mollick, and colleagues at Harvard, Wharton, MIT and Boston Consulting Group (BCG) described AI as a “jagged technological frontier.” These systems do striking work on one task while falling flat on something that looks just as easy. When a task sits outside what a model can do well, better direction alone may not close the gap. You can be an excellent driver at the wheel but you won’t make it up an unplowed hill. That’s why I put knowing when not to use it in the first lesson instead of an advanced one.
A second paper from that same research program watched 244 BCG consultants work. Some split the task, giving part to AI while keeping the rest for themselves. Others worked back and forth with the AI model. Meanwhile, some handed over both the work and judgement to AI. The first group was deepening their understanding of the work, the second group was building capability with the AI tool, but the third was building neither.
That third group made up 27% of the study. While the first two groups kept their hands on the wheel, the third simply handed the work over. Tiering doesn’t help there. You have to be in the driver’s seat or else the level of instruction means nothing.
What’s Left When the Tool Is Gone
Although driving has a clear curriculum, working with AI doesn’t. Context windows, cost, and the move toward agentic workflows have all shifted so fast that material taught even six months ago needs rebuilding. I don’t know of an institution that will keep pace with the full breadth of this area, including ours.
We need to build training around the driving. The mechanics matter, but they get swapped out every eighteen months.
Four things worth doing:
- Match the depth to the role. Select training and material for what the role must do.
- Teach the mechanics in service of the driving. Understanding the technical fundamentals behind AI can lead to more thoughtful use. Teach them where they change how someone executes.
- Give the reviewers reps too. Don’t just educate the users and technical staff. Managers, reviewers, and those setting direction for the team are also in need of strong AI skills.
- Test your training by varying or removing the tool. If something was taught with one model or tool, try it with another. See if the method carries over beyond AI or not. Portability of the skill demonstrates the strength of the training.
Learn More
To continue your AI driver’s ed, explore professional development courses from InterPro. Available training includes Artificial Intelligence: Strategy and Governance, and more!