Claude for L&D — Interactive Guide

Parker Lee · AI for Instructional Design

Claude for L&D — an interactive guide

What Claude is, what Skills and MCP unlock, and what you can actually build — for L&D folks who are curious but don't know where to start.

What is Claude?

A reasoning partner — not a search engine, not autocomplete

The mental model

Think of Claude as a brilliant colleague who has read everything but hasn't done your specific job. It knows ADDIE, Bloom's Taxonomy, action mapping, and xAPI — but it doesn't know your learners, your company culture, or your SME's particular quirks.

That division of labor is the whole game: you bring context, relationships, and judgment. Claude brings speed, pattern recognition, and tireless output.

What Claude IS

✓A reasoning engine — it can think through a problem, not just retrieve an answer
✓A writing accelerant — first drafts, rewrites, narration scripts, scenario branches in seconds
✓A structure builder — takes messy SME notes and returns a coherent outline
✓A code generator — builds interactive HTML tools, simulations, and learning apps
✓A tireless reviewer — critiques your work without getting bored on draft seven

What Claude ISN'T

✕Your domain SME — it will sound confident about things it's wrong about. Always get expert review on technical content
✕A source of truth — it doesn't know your organization, your learners, or last quarter's policy change
✕Autonomous (without tools) — by default it can only work with what you put in the conversation
✕A judgment replacer — when two design choices are genuinely in tension, that call is still yours

How it differs from other tools

vs. Search engines

Search retrieves existing content. Claude reasons through new problems.
Search gives you links. Claude gives you drafts.
Search doesn't know your constraints. Claude works within them.

vs. ChatGPT / other LLMs

Similar capability class — the key difference is the context window (how much it can hold at once) and instruction-following quality.
Claude handles very long documents well — full storyboards, entire policy PDFs, multi-course curricula.

A real exchange — what this looks like in practice

conversation

You → Here's a brief I got from a stakeholder: "We need training on our new returns policy — it changed last month." What questions should I ask before I design anything?

Claude → Before you design anything, I'd push on these: 1. What behavior is breaking down — are people processing returns wrong, or do they not know the policy changed? 2. What does a bad return interaction actually cost? (Returns, refunds, customer escalations?) 3. Is this a knowledge problem or a process problem? If the system doesn't surface the new policy at the moment of decision, training won't fix it. 4. Who are the learners — floor staff, phone agents, managers? Their context is completely different. 5. What's the actual deadline here, and does it allow for a pilot?

/skill

Skills

Pre-packaged workflows — expertise in a single command

What is a skill?

A skill is a saved workflow that encodes the right sequence of thinking steps for a specific task. Instead of writing a long, careful prompt every time you need a scenario draft or a needs analysis, a skill does that setup for you automatically.

In Claude Code (the desktop app), skills appear as slash commands: type /scenario-draft and Claude already knows the format, the questions to ask, and the output structure you want. In the Claude Agent SDK, skills are defined capabilities that agents can invoke as part of a larger workflow.

Why this matters for L&D

→Consistency — every scenario draft starts from the same structure, regardless of who on the team runs it
→Speed — no rebuilding the prompt from memory; the skill holds the pattern
→Shareable — a skill file is plain text. Your team can use the same one.
→Improvable — when a prompt pattern gets better, update the skill file once and everyone benefits

Building your own

Skills are plain text Markdown files. No coding required. If you can write a good prompt, you can build a skill.

In Claude Code, save a .md file to .claude/commands/ in your project. Give it a name. That name becomes the slash command.

The rule

Find the task you do 10+ times a project (usually scenario writing or objective drafting). Get your prompt working great once. Save it as a skill.

L&D skill examples

Command	What it does	Input
/needs-analysis	Structures a brief into a diagnosis — separates training problems from process/tool problems	Stakeholder brief or interview notes
/scenario-draft	Scaffolds a branching scenario with realistic trigger, 3 choices, and consequence-driven feedback	Situation description + wrong behavior to train against
/objective-writer	Converts vague goals into Bloom's-aligned, application-level learning objectives	Stakeholder's stated goal
/sme-interview	Generates action-mapping SME interview questions — behavior-focused, not knowledge-focused	Topic area + role of SME
/kirkpatrick-plan	Builds an L1–L4 evaluation plan with realistic data sources and attribution honesty	Project brief + available data sources
/launch-email	Drafts learner and manager launch communications that lead with learner benefit, not company need	Course summary + audience

What a skill definition looks like (simplified)

.claude/commands/scenario-draft.md

You are an instructional designer creating a branching scenario.

Given the situation the user describes:
1. Write a realistic opening trigger (2–3 sentences, present tense, specific setting)
2. Offer exactly 3 choices — none obviously correct, all plausible
3. For each choice, write the realistic consequence (not "try again" — what actually happens)
4. Write a debrief that explains the reasoning, not just the correct answer

Use conversational language. Avoid corporate training clichés.
Ask the user for the situation if they haven't provided one.

MCP

MCP & Tools

How Claude connects to the world outside the conversation

What is MCP?

MCP stands for Model Context Protocol. It's an open standard (published by Anthropic) that defines how AI models connect to external tools, data sources, and systems. Think of it as the universal connector between Claude and the rest of the software world.

Without MCP, Claude can only work with what you manually paste into the chat. With MCP, Claude can read files, search the web, write documents, call APIs, and run code — all within a conversation.

Without MCP

Sealed room

Claude is a brilliant advisor sitting in a room with no windows or doors. You have to carry every piece of information in yourself — copy-paste your SME notes, your storyboard template, your LMS report. Claude responds, and you carry the output back out manually.

With MCP

Connected workspace

MCP opens the doors. Claude can read your project files directly, pull data from Notion, search the web for compliance requirements, write files back to your hard drive, and run the code it just generated — all without you copy-pasting anything.

MCP tool categories for L&D

File system Read SME interview notes, write storyboard drafts, organize content libraries — directly from your project folder

Web search Research industry benchmarks, pull compliance requirements, find current data without leaving the conversation

Notion / Google Docs Read existing content docs, update project pages, sync resources across your workspace

Code runner Generate and immediately test HTML/JavaScript for interactive learning tools — Claude writes it, runs it, and fixes errors in one loop

Browser Preview what a course looks like in a browser, reference live UI for simulator design, capture screenshots

LMS / API Pull completion data, analyze assessment results, generate reports — if your LMS has an API, MCP can connect to it

What a connected workflow looks like

Claude with file system + web search tools

You → Read the SME interview notes in /project/sme-notes.txt, then generate a storyboard outline using our standard template at /templates/storyboard-template.md. Save the result to /project/storyboard-draft.md.

Claude → [reads sme-notes.txt] I can see 3 key behavior gaps in the notes. [reads storyboard-template.md] Using your template structure. [generates storyboard] [saves file] Done — storyboard saved to /project/storyboard-draft.md. I found 8 scenarios in the SME notes and drafted 3 of the highest-stakes ones in full detail. Want me to draft the remaining 5, or should we review the first batch first?

The shift

Without tools: Claude is a writing assistant. With tools: Claude is a workflow partner that can complete multi-step L&D tasks end to end — without you manually moving information between steps.

Build

Build with AI

A catalog of L&D artifacts Claude can help you create

Artifact catalog

◆ Discovery & Analysis

Needs analysis reports from interview notes
Performance gap diagnoses
SME interview question guides
Audience analysis summaries
Learning Design Documents (LDD)

◆ Course Content

Rise 360 content outlines with section structure
Storyline slide scripts & branching logic maps
Full narrator scripts (VO-ready)
Knowledge check questions with distractors
Scenario scripts with character dialogue

◆ Interactive Tools

HTML/JS learning simulations (no Storyline required)
Branching scenario players
Sales enablement product builders
Software process simulators
Decision-tree job aids

◆ Delivery & Evaluation

Manager one-pagers for launch prep
Learner launch communications
L1 reaction surveys
L3 behavioral observation checklists
Kirkpatrick evaluation plans

What to hand Claude vs. what to keep

Hand to Claude

First drafts of anything — especially things you'd revise anyway
Structural decisions (how to organize content, which modality to recommend)
Bulk generation (10 scenario branches, 20 quiz questions)
Plain HTML and JavaScript for interactive tools
Language polishing and consistency checks

Keep in your hands

Domain accuracy review — always have an SME check technical content
Learner empathy — Claude doesn't know who's actually sitting in front of the course
Judgment calls when two design directions are genuinely in tension
The relationships: sponsor, SME, manager — that's still human work

Full build example — HTML learning tool

Claude Code with file system tool

You → Build a de-escalation scenario trainer in plain HTML. The learner reads a frustrated customer message, picks a response from 3 options, and gets feedback with an empathy score. Save it to /output/deesc-trainer.html

Claude → [generates HTML + JS + CSS] [saves file] [tests in browser] [fixes a bug in the scoring logic] Done. The file is at /output/deesc-trainer.html. It includes 5 customer scenarios, scored responses on empathy + resolution speed, and a summary screen at the end. Want me to add a print-to-PDF button for use as a coaching artifact?

Go →

Start Here

Five days to working fluency with Claude

Your first week

1 Day 1 — Pressure-test a brief. Take a current project brief and paste it into Claude. Ask: "What questions should I ask before designing training for this?" Watch what it surfaces that you'd have missed. Don't use the output directly — use it to sharpen your own thinking.

2 Day 2 — Draft SME questions. Before your next SME interview, ask Claude to generate action-mapping questions for the topic. Compare them to what you'd have written. Use the best of both — you'll probably take 60% Claude, 40% yours.

3 Day 3 — Restructure your notes. After an SME interview, paste your raw notes and ask Claude to extract the behaviors that distinguish expert from novice performers. It won't be perfect — but it'll be 80% of the way there in 30 seconds.

4 Day 4 — Build something with code. Describe a simple interactive piece you wish you had: a decision-tree job aid, a talking-point flashcard, a quiz. Ask Claude to build it in plain HTML. Even if it's rough, you'll see what's possible.

5 Day 5 — Save your best prompt as a skill. Identify which task you repeated most this week. Write a prompt that produces the output you want. Save it in a text file. That's your first skill — you can formalize it later.

What to watch for

→When Claude sounds confident and is wrong — this is your signal to push for SME review. It always sounds confident.
→When the output is generic — you didn't give it enough context. Add your learner profile, your company context, your constraints.
→When it gives you a both-sides answer — that's Claude avoiding the decision. Push back: "Given these specific constraints, which would you choose and why?"

The only rule that matters

Core principle

Use Claude on real work, not practice work.

The fastest way to develop judgment about what Claude is good at is to see it succeed and fail on actual projects. Toy prompts don't build intuition. The moment it saves you an hour on something that actually mattered, you'll know how to use it.

The workflow that actually works

Use Claude to accelerate

Pressure-test the brief before you design anything
Draft interview questions, scenario scaffolds, objective language
Build interactive HTML tools you couldn't build alone
Review your own drafts for corporate-speak and unclear logic

Then filter hard

Take what's useful, discard the generic
SME-verify anything domain-specific
Apply your judgment to the calls Claude can't make
Save your best patterns as skills for next time