You’re being asked to develop next year’s learning and workforce capability plan.
Leadership wants to know how quickly employees can become AI-ready. Finance wants to know whether AI will reduce training costs. Business leaders want productivity improvements. Employees want to know which skills they should develop to remain valuable.
And you’re expected to provide answers while the technology, the work, and the expectations surrounding both are still changing.
If you invest heavily in today’s skills, some may become less relevant before the programs are fully deployed. If you wait for job requirements to stabilize, the organization risks falling behind. And if you focus on teaching people how to use the latest AI tools, you may produce employees who are comfortable with the technology but no better prepared to perform their actual work.
So how do you build a credible capability strategy when you cannot confidently predict what the work will look like?
The answer is not to abandon long-term planning. It’s to change what you’re planning around.
The problem with training for today’s jobs
Most learning plans begin with existing roles, current competency requirements, and identified skill gaps. That approach makes sense when the work itself is relatively stable.
But what happens when AI changes how the work is performed?
Consider an employee who previously spent several hours researching information, preparing a recommendation, and writing a report. With AI, the first draft might take minutes.
Does that mean the employee needs less capability? Not necessarily. They may now need to judge whether the recommendation is sound, spot where it is wrong, and explain a decision they did not build from scratch.
The task has changed. The standard of competent performance may need to change with it. Yet an organization could respond by simply teaching employees how to generate reports faster. The training would be relevant to the technology, but incomplete for the work.
The risk isn’t just training people on skills that become obsolete. It’s failing to develop the capabilities that become more important when the work changes.
I learned that the practical way, leading learning for a workforce of more than 60,000 people. The programs that held were the ones tested against real work first.
Start with the work that needs to be done
You don’t need to predict every job that will exist five years from now to make useful decisions today. You need to understand which important tasks are changing, what good performance looks like, and where employees are likely to struggle.
Take a customer service role, where AI can now generate a response almost instantly. The learning requirement is no longer just knowing where to find information or how to write a response. Employees must also recognize when the AI has misunderstood a customer’s situation, applied an incorrect policy, or produced an answer that sounds convincing but is wrong. A polite, complete answer can still be wrong, and that is exactly the case that tests whether someone is ready.
Instead of asking whether employees know how to use an AI assistant, we should ask whether they can resolve customer problems accurately and efficiently when AI is part of their workflow, and whether they can recognize a flawed response.
This example is hypothetical, but the planning principle applies across many roles.
Start with the performance the business needs. Then determine what people must be capable of doing in the redesigned work.
Don’t confuse familiarity with readiness
One of the easiest mistakes is to measure AI readiness through participation. Employees attend an AI workshop. They complete a course. They demonstrate several prompts. They receive a certificate. The dashboard suggests progress.
But can they use AI effectively when the situation is unfamiliar, the information is incomplete, or the consequences of an error are significant? That requires a different kind of evidence.
Give employees realistic cases, including situations where the AI produces an incorrect or incomplete answer. Observe whether they identify the problem and explain their reasoning. Then examine the operational results: are customer issues resolved more quickly, and has rework increased because employees are accepting incorrect AI-generated answers?
These measures reveal much more than course completion. They also expose problems training alone cannot fix, such as unreliable information, slow verification, or ambiguous policy. A credible learning strategy must recognize that distinction rather than treating every performance problem as a training need.
Build a plan that can change as the work changes
The traditional instinct is to create a comprehensive learning roadmap, establish the curriculum, secure the budget, and roll out the programs. But when job requirements are evolving, a detailed plan can become outdated before it is fully implemented.
That doesn’t mean abandoning structure. It means building a different planning discipline.
Begin with a small number of business-critical workflows where AI is already being introduced or seriously considered. Establish how those workflows currently perform, then identify what AI might change, including the new judgments and responsibilities employees will need.
Then test the redesigned work with a small group before developing learning programs for everyone. That test should answer practical questions:
- Can employees perform the redesigned task reliably, including when AI makes mistakes?
- What knowledge and judgment do they need that existing training does not develop?
- Which problems require training, and which require changes to the technology or workflow?
- What evidence would justify expanding the approach?
Use the findings to improve the workflow, define the required capability, and build learning around demonstrated gaps.
The employees involved in those early tests can also become valuable guides for others. But their contribution should go beyond promoting the change. They need opportunities to identify what doesn’t work and help improve it.
Once the approach produces credible results, expand it. This is not an argument against long-term capability planning. It’s an argument for making long-term plans responsive to evidence rather than dependent on predictions that may not hold.
Address the question employees are already asking
Employees are being asked to learn technologies that may fundamentally change their jobs. Some are understandably worried that becoming more productive could eventually make their positions unnecessary.
Telling them to embrace AI or focus on higher-value work doesn’t resolve that concern. What higher-value work? Which capabilities will matter? How will the organization make decisions about changing responsibilities?
Learning leaders may not control staffing decisions, but they can help leadership answer the capability questions those decisions depend on. When work is redesigned, identify what responsibilities are being removed, what new responsibilities are emerging, and where employees will need support. Where the future remains uncertain, acknowledge it rather than promising outcomes the organization cannot guarantee.
A better question for your next planning meeting
The next time someone asks for your AI learning roadmap, resist starting with a list of courses, platforms, or certifications.
Start with one important question:
Which business-critical work is changing, and what will our people need to do differently to perform it successfully?
Choose one role or workflow. Define what good performance should look like with AI involved. Test those requirements against actual work. Use the evidence to decide what to develop and what to redesign. Stop what the evidence does not support.
The objective isn’t to prepare people for every possible future. It’s to help them perform effectively as the work changes, while building the capacity to adapt again.
That is a capability strategy worth investing in. The open test is which workflow you will start with.
Dr. Ravinder Tulsiani


