Skip to main content

Teacher’s Edition

Module 11

Claude Code for Non-Engineers

What becomes possible when code can be created on demand. A data pipeline in five minutes. An internal tool in fifteen. When non-engineers watch a working tool get built in the time it takes to file a Jira ticket, something shifts.

70 minutesAll Tier III participantsPrerequisite: Module 10

Charter Oak Strategic Partners · Claude Mastery Program · Version 1.0 · Confidential · Not for distribution to participants

Back to Module|Teacher’s Guide
Concept: What Claude Code Actually Is

Claude Code is a command-line tool that runs in a terminal. It reads files, writes files, edits code, runs shell commands, searches codebases, and manages git operations. Developers use it to build software. This module is not teaching participants to become developers. It is teaching them to recognize what becomes possible when software can be created on demand by describing what you want in plain English.

The concrete difference from Cowork: Cowork creates documents and reports from files on your computer. Claude Code creates software — scripts, tools, applications, automations, dashboards. Cowork is your knowledge work colleague. Claude Code is your on-demand engineering team.

The target audience for this module is the operations manager who waits six weeks for IT to build a reporting tool, the sales lead who maintains a 47-tab spreadsheet because nobody has time to build a real dashboard, and the HR director who manually processes onboarding documents for every new hire. When they see Claude Code build a working tool in the time it takes to write the Jira ticket, their understanding of what is achievable shifts permanently.

Concept: Claude Code vs. Cowork: When to Use Which

This question will surface immediately. Have the answer ready:

Use Cowork when the output is a document, a report, a presentation, an analysis, or file-based knowledge work. Cowork reads existing files and creates new ones. It is optimized for the kind of work that ends in a PDF, a Word doc, an email, or a slide deck.

Use Claude Code when the output is a tool, a script, an automation, an application, or a data pipeline. Claude Code writes code, runs it, tests it, and iterates until it works. It is optimized for the kind of work that ends in software someone can run repeatedly.

The overlap: Cowork can write and execute code inside its sandbox, and Claude Code can create documents. The distinction is about the primary output. If you are building something that will be run, reach for Claude Code. If you are building something that will be read, reach for Cowork.

Opening: Why This Matters to Non-Engineers — 10 minutes

Script: The Opening Frame

“Raise your hand if you have ever waited more than a month for a tool or report that IT was supposed to build.” Hands go up. “Keep your hand up if the tool you eventually got was not quite what you asked for.” Most hands stay up.

“That gap — between what you need and what engineering has time to build — is the reason this module exists. Claude Code does not replace your engineering team. It fills the gap between ‘I need a tool’ and ‘engineering will get to it in Q3.’”

“I am going to build three working tools in front of you in the next thirty minutes. None of them require you to understand code. All of them solve problems you probably have right now.”

demo-data/module-11/claude-code-scenarios.mdThree scenarios with prompts, expected output, and the decision matrix.

Demo 1: Data Pipeline — 5 minutes

Script: The CSV Reformatter

Open a terminal. Type claude to launch Claude Code. The audience sees a command-line interface with a text prompt, not a chat window. This visual difference is intentional — it signals that this is a different kind of tool.

Type: “Build a Python script that reads quarterly-sales-messy.csv, cleans up the column headers (remove spaces, lowercase everything), fills in missing region values by looking at the previous row, and outputs a clean CSV called quarterly-sales-clean.csv.”

Narrate what the audience sees: “Claude Code is reading the file. It is writing a Python script. It is running the script to test it. It found an edge case — some rows have ‘N/A’ instead of blanks — and it is fixing the script. Now it is running it again. Clean file created.”

Open both files side by side. Show the messy input and the clean output.

“Five minutes. This is the task someone on your team does manually every week with find-and-replace and conditional formatting. The script is reusable. Run it next week with a new file, same result.”

Facilitator Note

If you are not comfortable running Claude Code live, pre-record a screen capture of the three builds and narrate over the recording. The key teaching moment is the speed and the working output, not the live-typing experience. A pre-recorded demo with confident narration is better than a live demo with fumbling.

Demo 2: Internal Tool — 15 minutes

Script: The Customer Visit Scheduler

“This one is more ambitious. I am going to ask Claude Code to build a tool that a sales rep can use to prepare for customer visits.”

Type: “Build an HTML page with a form where a sales rep enters: customer name, visit date, visit objectives, and key contacts. When they click Generate, the page should create three things: a formatted visit brief, a draft calendar invite text, and a pre-visit checklist. Use clean, professional styling. No external dependencies — everything in one HTML file.”

Narrate the build process. Claude Code will create the HTML file, test it, and iterate on the design. Open the finished page in a browser. Fill out the form. Click Generate. Show the three outputs appearing on the page.

“Fifteen minutes. This is the kind of tool that sits on an IT backlog for six months because it is useful but not urgent enough to prioritize. Claude Code builds it in a quarter-hour. The sales rep uses it tomorrow.”

Watch For: The “But Is It Real?” Question

Someone will ask whether this is production-quality code. The answer is no, and that is the point. This is not meant to replace professionally engineered software. It is meant to replace the spreadsheet workaround, the manual process, and the tool request that never gets built. The quality bar is: does it work? Is it better than what we have now? If the answer to both is yes, it has value.

Frame it explicitly: “Production systems go through engineering. Internal tools, personal automations, and prototypes can come from Claude Code. The distinction is: who uses it and what are the consequences if it breaks?”

Demo 3: Automation Script — 10 minutes

Script: The Multi-Report Dashboard

“Last one. Three CSV files: sales data, expense data, headcount data. I want a single dashboard that shows key metrics from all three, with charts, updated by running one command.”

Type: “Build a dashboard that reads sales.csv, expenses.csv, and headcount.csv. Calculate: total revenue, total expenses, revenue per employee, expense ratio, and month-over-month changes. Create an HTML dashboard with charts for each metric. Use Chart.js from a CDN for the charts. Save as dashboard.html.”

Show the finished dashboard in a browser. Click through the charts. “One command. Updated dashboard. This replaces the Friday afternoon ritual of copying numbers from three spreadsheets into a PowerPoint.”

The Four-Interface Decision Matrix — 15 minutes

Concept: The Complete Claude Ecosystem

With three demos done, formalize the framework. Four interfaces, four use cases. This is the mental model participants take back to their desks:

Claude Chat. Best for: quick questions, drafting text, brainstorming, one-off analysis. The input is a question or a prompt. The output is text. No file access, no persistence, no execution. Use it when you need one good answer to one question.

Claude Projects. Best for: persistent context, team knowledge bases, ongoing conversations about a body of work. Upload documents, set custom instructions, have focused chats with Claude that reference the full project context. Use it when you need Claude to understand a large body of material across multiple conversations.

Claude Cowork. Best for: file-based knowledge work, document creation, multi-step analysis, report generation. Cowork reads and writes files from your computer, runs code in a sandbox, creates output documents, and can run for extended periods. Use it when the deliverable is a document, a report, or an analysis that involves local files.

Claude Code. Best for: building tools, writing scripts, creating automations, data pipelines, prototyping applications. Claude Code runs in a terminal, writes and executes code, manages files, and iterates until the tool works. Use it when the deliverable is software someone will run.

Claude API. Best for: production systems at scale, integrating Claude into existing software, processing hundreds or thousands of requests programmatically. Use it when humans are not in the loop for every interaction. (Module 13 covers this in depth.)

Script: Testing the Matrix with the Room

“I will give you a task. You tell me which interface. Ready?”

“Task: You need to rewrite a client email to sound more professional.” Answer: Chat. It is a quick, one-off text transformation.

“Task: You need to build a script that reformats your CRM export every week.” Answer: Claude Code. It is a reusable tool.

“Task: You need to analyze three contracts for inconsistent terms.” Answer: Cowork. It is multi-file analysis that produces a document.

“Task: You want every support ticket automatically categorized and routed to the right team.” Answer: API. It runs at scale without a human.

“Task: Your marketing team needs ongoing access to brand guidelines and competitive research when chatting with Claude.” Answer: Projects. It is persistent background context.

Do six to eight scenarios. Let tables discuss before revealing the answer. The wrong answers are as instructive as the right ones — they reveal where the boundaries feel ambiguous.

Ask the Room: Boundary Cases

“What about a task that could use either Cowork or Claude Code?” Good question. If the output is primarily a document that happens to involve some data processing, start with Cowork. If the output is primarily a script or tool that produces documents as a side effect, start with Claude Code. When in doubt, try Cowork first — it is more accessible for non-engineers.

Group Discussion — 15 minutes

Ask the Room: Application Questions

“Which of the three demos would save the most time in your organization? Be specific. Name the team, name the task, estimate the hours.”

“Who on your team should learn Claude Code? It does not have to be an engineer. Who has the right combination of technical curiosity and domain knowledge?”

“What is on your IT backlog right now that could be built in an afternoon with Claude Code?”

Ask the Room: Risk Discussion

“What are the risks of letting non-engineers build tools with Claude Code?” Surface these explicitly:

Data security. A Claude Code script can read and process sensitive files. Who reviews what data the script accesses?

Maintenance. If the person who built the script leaves, who maintains it? Code that nobody understands is a liability.

Code quality. Claude Code writes functional code, but it may not follow the security standards, error handling, or logging practices your engineering team requires.

Over-reliance. If a critical process depends on a Claude Code script, what happens when it breaks? Is there a fallback?

The goal is not to scare participants away from Claude Code. It is to give them the vocabulary to have a productive conversation with their technical team about boundaries.

Debrief — 10 minutes

Script: The Boundary Between Build and Engineer

“Claude Code builds things fast. Speed is not the hard problem. The hard problem is knowing what to build with Claude Code and what should go through your engineering team.”

“Here is the line: internal tools, personal automations, prototypes, data pipelines, one-off scripts — Claude Code territory. Production systems, customer-facing features, anything touching sensitive data at scale, security-critical code — engineering territory.”

“The highest-value use of this module is giving you the ability to prototype what you need and show it to your engineering team. Instead of writing a requirements document that gets misinterpreted, you hand them a working prototype and say: ‘Make this production-ready.’ The conversation changes completely.”

Concept: The Claude Agent SDK (Advanced Context)

For participants who want to go deeper: Anthropic provides a Claude Agent SDK that allows developers to embed Claude Code’s capabilities into their own applications. The SDK supports custom system prompts, tool permissions (Read, Write, Edit, Bash), and integration with project-specific instructions via CLAUDE.md files. This is how companies build Claude-powered internal tools that teams use without opening a terminal.

This is Module 13 territory (API) and does not need to be taught here. Mention it as context for participants who ask “how do we scale this?” The answer: the SDK turns Claude Code’s capabilities into building blocks for production applications.

SegmentActivityTime
OpeningWhy non-engineers need to see this10 min
Demo 1Data pipeline: CSV reformatter5 min
Demo 2Internal tool: customer visit scheduler15 min
Demo 3Automation: multi-report dashboard10 min
FrameworkFour-interface decision matrix15 min
DiscussionApplication, risk, and boundary questions15 min
DebriefBuild vs. engineer boundary10 min