1.2

What AI Can (and Can't) Do

Realistic expectations for your work

✅ Capabilities ⏱️ ~15 minutes

Let's Get Specific

You know AI is changing the game. Now let's talk about exactly what it can and can't do for instructional design work. No marketing hype, no dystopian fear-mongering—just honest capabilities based on current technology.

The key to using AI effectively: Match the task to AI's actual strengths. When you ask AI to do what it's genuinely good at, the results are remarkable. When you ask it to do things it struggles with, you'll be disappointed.

What AI Does Exceptionally Well

1. Generating First Drafts

AI excels at creating initial versions of content you can refine. This includes:

  • Learning objectives aligned with Bloom's taxonomy
  • Course outlines and module structures
  • Scenario-based questions
  • Multiple-choice assessment items
  • Facilitator guide scripts
  • Explanatory content on specific topics
Real example:
"I asked Claude to write 10 learning objectives for a customer service training. It gave me 10 properly formatted objectives using action verbs. Were they all perfect? No—two were too vague, one was redundant. But editing 10 objectives takes 20 minutes. Writing 10 from scratch takes 2 hours."

2. Brainstorming and Ideation

AI is tireless at generating options. Need scenario ideas? Assessment formats? Engagement strategies? AI can produce 20 variations in seconds.

3. Format Conversion

AI can transform content between formats easily:

  • ILT workshop → eLearning module
  • Long-form course → microlearning series
  • Text-heavy content → table or infographic structure
  • Technical documentation → learner-friendly explanation

4. Pattern Recognition and Application

Once you show AI what "good" looks like in your context, it can replicate that pattern across new content. Provide 3 examples of your organization's scenario style, and AI can generate 20 more in the same voice.

5. Content Summarization

AI can distill long documents into key points, extract main ideas from SME interviews, or condense research into learner-appropriate content.

What AI Does Reasonably Well (With Oversight)

1. Creating Realistic Scenarios

AI can generate plausible workplace scenarios, but you need to check for:

  • Organizational accuracy (it won't know your specific procedures)
  • Appropriate difficulty level
  • Authentic dialogue and decision points

2. Building Assessments

AI creates decent assessment items, but you must verify:

  • Distractors are plausible but clearly wrong
  • Questions test actual learning, not trivia
  • Difficulty aligns with objectives
  • No bias or trick questions

3. Generating Visual Descriptions

AI can create prompts for image generation or describe what visuals would support content, but the actual images from AI generators (DALL-E, Midjourney) require iteration and often manual cleanup.

What AI Struggles With

1. Organizational Context

AI doesn't know:

  • Your company's specific terminology or processes
  • Your learner population's background knowledge
  • Your organization's culture or brand voice
  • Compliance requirements for your industry

💡 The Context Gap

AI generates "generic professional" content by default. To get content that feels authentic to your organization, you must explicitly provide that context in your prompts.

2. Instructional Strategy Decisions

AI can suggest approaches, but it can't reliably determine:

  • Whether this topic needs practice activities or just information
  • If learners should discover concepts or receive direct instruction
  • What level of scaffolding your specific audience requires
  • Which engagement strategies will resonate with your learners

3. Quality Judgment for Your Standards

AI can identify grammatical errors and generic quality issues, but it can't evaluate:

  • Whether content meets your organization's style guide
  • If examples are culturally appropriate for your audience
  • Whether difficulty progression is appropriate
  • If the tone matches your brand

4. Emotional Intelligence and Sensitivity

AI misses nuance in:

  • Sensitive topics (DEI, health issues, workplace conflicts)
  • Appropriate humor or lightness
  • Reading the room (knowing when to be formal vs. casual)
  • Cultural considerations beyond obvious stereotypes

What AI Simply Can't Do

1. Conduct Needs Analysis

AI can't interview stakeholders, observe workplace performance, or identify the real problem behind a training request.

2. Make Strategic Decisions

Is training even the right solution? Should this be eLearning or ILT? Is this a knowledge problem or a motivation problem? These require human judgment.

3. Navigate Politics

AI won't help you negotiate with difficult SMEs, manage stakeholder expectations, or present to executives.

4. Guarantee Accuracy

AI confidently generates plausible-sounding content that can be completely wrong. It "hallucinates" facts, especially about:

  • Specific statistics or research findings
  • Technical procedures it hasn't been trained on
  • Recent events (knowledge cutoffs apply)
  • Niche or specialized topics
Critical rule:
Never publish AI-generated content without fact-checking. If accuracy matters (and it almost always does), verify everything against reliable sources.

The Sweet Spot: Where to Start

Based on current capabilities, here are the ID tasks where AI provides the best return on effort:

  1. Drafting learning objectives — Fast, high success rate, easy to refine
  2. Creating practice scenarios — Good starting points, requires context and editing
  3. Generating assessment items — Produces quantity, you ensure quality
  4. Outlining course structures — Helps overcome blank page syndrome
  5. Adapting content to different formats — Mechanical task, reliable results

🎯 Start Here

For your first AI experiment, choose task #1 or #4 from the list above. These have the highest success rate and lowest risk. Once you've seen AI work well on something simple, you'll have more confidence tackling complex tasks.

Managing Expectations

Here's a realistic expectation framework:

For routine, well-defined tasks: Expect 60-80% of AI output to be usable with minor edits. Your time savings: 50-70%.

For complex, context-dependent tasks: Expect 30-50% of AI output to be usable. Your time savings: 20-40%.

For specialized or sensitive content: Expect AI to provide ideas and structure, but plan to rewrite substantially. Your time savings: 10-25%.

These aren't failures—20% time savings on complex work is still significant when you're juggling multiple projects.

Looking Ahead

Now that you know what AI can realistically do, the next topic covers the specific tools available: Claude, ChatGPT, DALL-E, Midjourney, and others. You'll learn which tool is best for which task, and why you might use different AI tools for different parts of your workflow.