Bank of America is building generative AI skills around a basic operating principle: employees remain responsible for the work AI produces. For a bank with more than 200,000 teammates, that principle shapes how its onboarding and professional development group, The Academy, introduces the technology. The scale matters because a training approach used across a large workforce has to do more than explain how to write a prompt. It has to give employees a shared way to think about when an AI answer is useful, what needs checking, and who owns the result.
Michael Wynn, the Academy’s Enterprise Learning and AI Strategy Executive, sums up the approach this way: “AI is the car, and you are the driver.” The metaphor puts the employee in control of the tool and accountable for its output. That distinction can sound obvious, but it addresses a practical risk: a fluent answer can appear finished even when it needs correction, added context, or a closer review. The system may help produce the work; it can't take responsibility for whether that work is fit for its purpose.
Thinking before prompting
Training starts with a pen and paper. New teammates write down the result they want, the relevant context, and how they’ll judge whether an answer is useful. Wynn describes this as a way to recognize “the value of slowing down to think before you prompt.” The exercise develops problem definition and critical thinking before employees turn to prompts, putting the task ahead of the tool.
That sequence has a practical logic. A prompt can only guide a system toward the information and framing it receives, so an employee who hasn't settled the goal may get an answer that's polished but poorly matched to the need. Writing down the desired result and the criteria for judging it gives the user a point of comparison. Without that step, it can be tempting to treat the first plausible response as a successful one.
Here's the thing: prompt writing is only part of the skill. Employees also need to notice when the response leaves out a relevant condition, makes an unsupported claim, or answers a different question from the one they meant to ask. A written plan doesn't guarantee a correct result, but it can make gaps easier to spot. And it gives the employee a reason to revise or reject an answer instead of accepting it on presentation alone.
Employees remain accountable
The Academy begins with strategic prompt design workshops and responsible AI use training. Its introductory prompting course establishes a “human in the lead” standard: teammates must review, edit, and approve AI-generated work. The steps matter because review is an active part of producing the final work, not a quick check after the system has already shaped the answer.
Similar systems typically produce responses from the instructions and context they receive; they don't know, on their own, whether an answer meets a workplace need. That leaves employees with several distinct tasks: checking whether the response addresses the request, judging whether its details are supported, and deciding what changes are needed before use. Bank of America’s stated standard assigns those decisions to the person using the tool.
(And this is the part most people overlook.) Accountability depends on employees having enough time and judgment to review what the technology produces. A requirement to approve an answer has limited meaning if approval becomes a routine click. The Academy’s emphasis on thinking before prompting gives review a stronger starting point: employees first define what they need, then have a basis for deciding whether the system delivered it.
Workplace training fills a gap
With formal generative AI instruction still absent from many academic curricula, workplace training often provides new hires with their first structured guidance. Bank of America’s approach makes clear that effective use depends on sound judgment before and after a prompt is entered. For employees encountering these tools at work for the first time, that guidance can set expectations about both their capabilities and their limits.
The company’s approach also reflects a broader training challenge. Teaching people to enter prompts is relatively concrete; teaching them to assess the answer requires practice with the underlying task. Employees need to know what a useful result looks like in their own work before they can reliably distinguish it from one that merely sounds convincing. That takes more than familiarity with a new interface.
If you're working in this space, the useful question is whether training makes employees better at deciding what to trust and what to change. The Academy’s model puts that judgment at the center, with prompt design as one step in a larger process. The outcome will depend on how consistently employees can apply the standard in their day-to-day work. A principle, by itself, doesn't show how well that happens.
What the approach signals
This is more significant than it looks: assigning responsibility to the employee changes the way AI training is framed. The focus isn't simply on helping staff produce answers faster; it’s on keeping human judgment attached to the work. That may make the program less flashy than a course centered on prompts, but it addresses a more consequential question for an organization using AI across a large workforce: who decides that an output is ready?
There are limits to what can be concluded from the description. It establishes the Academy’s training principles and the sequence of its introductory instruction, but it doesn't provide results showing how employee performance has changed. That distinction matters. A clear standard is a starting point; evidence of consistent review would show how effectively it carries into practice.
For now, the message is straightforward. AI can assist with work, but employees remain responsible for the result. Useful training, on this account, begins before the prompt and continues after the answer arrives. Then the judgment call.
The post AI is the car, and you are the driver: Bank of America’s approach to AI learning appeared first on Tearsheet.
