Stop Asking “What Can AI Do?” Start With the Learning Goal

AI has made it remarkably easy to start an instructional design project in the wrong place. If you open almost any AI tool, you can immediately generate learning objectives, outlines, scenarios, quiz questions, scripts, graphics, summaries, feedback, and even complete courses. If you give the tool a topic and a target audience, within seconds it will give you have something that looks surprisingly close to instructional content. While that speed is useful. It can easily distract us from a much more important question.

Before asking “What can AI do here?”, instructional designers should be asking:

“What do people need to learn or be able to do, and what thinking do they need to practice to get there?”

That one simple question is important because it changes how you use AI.

Instead of looking for tasks you can automate, you start deciding which parts of the work AI can reasonably support and which parts people need to do themselves. Sometimes AI should generate the first draft, sometimes, it should critique what a learner created, sometimes it should provide alternatives or simulate a conversation, and sometimes the best design decision is to leave AI out of the activity completely.

Here is a practical way to make those decisions.

The Problem With Starting With the Tool

Imagine that you have been asked to create training for newly promoted supervisors. Your organization has recently purchased access to several AI tools. Someone suggests adding an AI coach to the training. Another person wants AI-generated scenarios. Someone else suggests using AI to provide personalized feedback.

While all of those ideas are technically possible, none of them tells you whether AI belongs in the training. Start instead with the performance problem. Perhaps new supervisors struggle with difficult conversations. They know the policies, but have trouble responding when an employee becomes defensive, provides incomplete information, or challenges their decision. Now you have something you can design around. The goal is for supervisors to make sound decisions during difficult conversations and explain those decisions appropriately. Once that is clear, you can evaluate AI based on whether it helps people develop that capability.

Step 1: Define What People Need to Do

Before opening an AI tool, complete this sentence:

After this learning experience, participants should be able to __________.

Be specific. For example, “Understand performance management” gives you very little direction, but “Conduct a performance conversation, respond appropriately to employee concerns, and document the agreed-upon next steps” gives you considerably more. The second version identifies observable performance. It also gives you clues about what learners need to practice.

For example, they may need to:

• recognize when an employee’s response requires additional questioning;
• decide which information matters;
• choose an appropriate response;
• explain their reasoning;
• document the conversation accurately.

Now AI has a defined role to support, rather than a blank space to fill.

This sounds like basic instructional design, and it is. The difference is that generative AI makes skipping this step unusually tempting because it can produce polished materials before the problem has been clearly defined.

Step 2: Identify the Thinking Behind the Performance

Next, ask a question instructional designers sometimes overlook:

What thinking does someone need to do to perform this task well?

Suppose you are designing cybersecurity training. A traditional objective might say: Employees will identify phishing emails, but what does that actually require? An employee may need to notice an unusual sender address, interpret the context of the request, recognize pressure tactics, compare the message against normal organizational practices, and decide whether to click, respond, report, or delete. Those decisions are the substance of the performance. Now imagine an AI assistant automatically analyzes every suspicious email and tells employees whether it is safe. That may be an excellent workplace tool, but it may be a poor learning activity if your goal is to teach employees how to recognize suspicious messages themselves.

The distinction matters because a tool can improve performance while reducing practice. This is why instructional designers need to know which one they are designing for.

Step 3: Decide What Learners Should Practice Themselves

Once you have identified the thinking behind the task, decide which parts learners need to perform without AI assistance. Ask: If AI does this for the learner, what does the learner no longer have to think about? That question is useful because AI often removes effort.

There is little educational value in requiring a learner to spend 20 minutes formatting a document if formatting has nothing to do with the learning objective. The decision becomes harder, however, when AI removes effort that is closely connected to the skill being developed.

Consider a leadership development program in which participants analyze a workplace conflict.

You could ask AI to:

• summarize the conflict,
• identify the stakeholders,
• identify possible causes,
• generate three solutions,
• compare the solutions,
• recommend the best option, and
• draft the manager’s response.

At that point, what is left for the participant to do? Perhaps read the answer and agree with it. The activity still looks sophisticated and the learner may even produce an excellent final response, but much of the reasoning the activity was designed to develop has already been completed.

A better design might use AI differently. Give learners the case and ask them to analyze the situation and make an initial recommendation. Then let them use AI to generate an alternative interpretation. Now the learner has something to evaluate:

What did the AI notice that I missed? Where do I disagree? Which recommendation is stronger? What evidence supports my decision? At that point, AI becomes part of the intellectual work without replacing it.

Step 4: Separate Productive Effort From Unnecessary Effort

Not all effort contributes equally to learning and this is where instructional designers need to make careful distinctions. Suppose employees are learning to write effective project proposals. They might spend time:

• researching background information;
• organizing ideas;
• developing an argument;
• evaluating evidence;
• formatting headings;
• correcting grammar;
• rewriting repetitive sentences;
• checking whether their recommendation addresses the business problem.

Some of those tasks may be central to the learning goal, but others may not be. If the objective is to develop strategic reasoning, AI assistance with formatting or proofreading probably does little harm.

Having AI determine the recommendation, select all the evidence, and construct the argument is different. Those activities may represent exactly what participants are supposed to learn.

Here’s a useful planning question:

Where does the learning actually happen in this task?

An instructional designer should protect that part and AI can often handle the rest.

Step 5: Decide What AI Should Do

Only now should you ask the question everyone wanted to ask at the beginning:

What can AI do here? At this stage, you have enough information to answer it intelligently.

AI might:

• Generate variety – Creating 20 realistic practice situations manually takes time. AI can help generate variations involving different roles, circumstances, levels of complexity, or constraints. An instructional designer can review and revise those scenarios before learners see them.
• Provide another perspective – Ask learners to make a decision first and then use AI to challenge it.

For example: I recommended Option B. Identify three weaknesses in my reasoning and one situation in which Option A might be more appropriate.

Now AI creates productive friction rather than completing the assignment.

• Simulate practice

AI can play the role of a customer, employee, patient, client, stakeholder, or manager. A learner can practice asking questions and responding to changing information without requiring another person to be available for every practice session.

• Generate a first draft

In professional learning, drafting may not be the skill you are trying to teach. AI could produce a first draft that learners have to evaluate, correct, or improve. In some cases, evaluating a flawed AI-generated response may actually be more useful than writing from scratch.

• Provide feedback

AI can provide immediate formative feedback when appropriate, especially during repeated low-stakes practice. But the instructional designer still needs to determine what the feedback should evaluate, what standards it should use, and when a qualified person needs to review the work.

Step 6: Build the Assessment Before You Get Too Attached to the AI

There is a simple test for whether your AI-supported activity makes sense: How will you know whether the learner can actually perform the skill? Let’s return to the supervisor example. Suppose participants spend an hour working with an AI coach that suggests responses throughout a simulated employee conversation, and perform beautifully. What have you demonstrated? You know they can perform the task with an AI coach providing assistance. That may be enough if supervisors will always have access to that type of support in the real workplace. But if they will need to conduct difficult conversations independently, you still need evidence that they can do so without the tool.

The assessment should match the conditions under which people will actually perform.

You might therefore design the experience in stages:

• Practice 1: Learner works through a scenario independently.
• Practice 2: Learner uses AI to examine alternatives and receive feedback.
• Practice 3: Learner handles a more complex scenario with limited assistance.
• Final assessment: Learner responds independently and explains the reasoning behind key decisions.

Now AI supports development without becoming the evidence of competence.

Step 7: Ask Who Is Making the Final Decision

There is another question worth adding to your design process: Where does professional judgment need to remain? This becomes especially important when AI is involved in decisions that affect people. Consider AI-generated feedback on employee performance, automated recommendations about learner progress, or an AI system that determines which employees need remedial training. The issue is not simply whether the system produces accurate output most of the time.

Someone still needs to decide:

• whether the recommendation makes sense in context;
• whether important information is missing;
• whether the system’s assumptions are appropriate;
• whether the learner or employee can challenge the decision;
• who is responsible if the recommendation is wrong.

Instructional designers may not own every one of these decisions, but we should be asking the questions.

A Simple AI Design Check

Before adding AI to a learning experience, work through these seven questions:

1. What should learners be able to do at the end?
2. What thinking or judgment does that performance require?
3. Which parts do learners need to practice themselves?
4. Which effort contributes little to the learning goal?
5. Where could AI provide useful assistance, variety, feedback, or challenge?
6. What evidence will demonstrate that meaningful learning occurred?
7. Where does a person need to retain responsibility for the final decision?

If you cannot answer the first three questions, you probably are not ready to answer the fifth.

What This Looks Like in a Real Design Workflow

The practical change is small, and I don’t recommend abandoning ADDIE, backward design, action mapping, performance consulting, or whatever process already works for you. You simply need to change the order of the conversation. Instead of: We have an AI tool, what can we create with it? Try: We have a performance problem, what capability needs to change? What practice will develop that capability? →Where could AI help? That sequence keeps the technology in its proper place. It also makes it easier to say no. Sometimes you will work through the questions and decide that AI can make an activity substantially better, but other times you may realize that adding AI would remove exactly the practice learners need.

The Question Instructional Designers Need to Keep Asking

AI will continue to make more instructional tasks easier. Course outlines that once took hours can be generated in minutes. Scenario development can happen almost instantly. Videos, images, assessments, simulations, feedback, and learner support are becoming easier to produce. The availability of those capabilities does not tell us when to use them.

That still remains a design decision.

For instructional designers, the useful question is no longer simply “Can AI do this?” We need to ask: If AI does this, what are people still learning to do? That question is significantly harder, but it also where instructional design matters most.