--- title: "Coding help and Claude Code, for non-engineers and builders" description: "Two ways Claude helps with code 1. Coding help in chat. Anyone can ask Claude to write a spreadsheet formula, a Google Apps Script, a small Python…" url: https://optimizeall.com/learn/mastering-claude/coding-help-and-claude-code updated: 2026-10-05 --- Mastering Claude (Anthropic) · Claude Code, connectors and Skills · lesson 13 of 20 · 16 min # Coding help and Claude Code, for non-engineers and builders ## Two ways Claude helps with code **1. Coding help in chat.** Anyone can ask Claude to write a spreadsheet formula, a Google Apps Script, a small Python script, a regex, or to explain an error message. Claude can also run code in a sandbox (code execution) to test ideas and process files. **2. Claude Code.** Anthropic's **agentic coding tool**. Instead of copying snippets back and forth, Claude Code works directly in a codebase: it reads files, plans changes, edits multiple files, runs commands and tests, and uses git, asking for your permission according to the permission mode you choose. It runs in the terminal, in IDEs such as VS Code and JetBrains, in the Claude desktop app, on the web (including cloud sessions you can check from your phone), and in Slack. It is included with paid Claude plans and is also available through the API. ## For non-engineers: the safe snippet pattern ```text I'm not a developer. I use Google Sheets. Goal: in column D, show "Overdue" if the date in column C is before today and column E is not "Paid"; otherwise leave blank. Give me: the formula, a plain-English explanation of each part, and 3 test rows I can type in to check it works (including an edge case). ``` For scripts that change files or data: **always test on a copy** of a small sample, read Claude's explanation, and keep backups. Never paste passwords or API keys into prompts or scripts; use the platform's secret storage. ## For builders: getting started with Claude Code Install it using the official quickstart at Anthropic's Claude Code documentation (native installer or package manager), then open your project folder: ```bash cd my-shopify-theme claude ``` Useful first moves: - Run `/init` to generate a **CLAUDE.md** file: project-specific guidance Claude reads at the start of every session (build commands, conventions, what not to touch). - Ask for a **plan first**: "Read the codebase and propose a plan to add a size-guide modal to product pages. Don't edit yet." - Work in **small, testable tasks** with a clear definition of done. - Review diffs before accepting; commit often so you can roll back. A minimal CLAUDE.md: ```markdown # Project: Noor Organics storefront - Stack: Shopify Liquid theme + small Node scripts in /scripts - Run checks: `npm run lint` and `npm test` before saying a task is done - Never edit files in /config/production or commit secrets - Arabic (RTL) and English layouts must both be checked - Keep changes minimal; explain any new dependency before adding it ``` ## Extending Claude Code - **MCP servers** connect Claude Code to tools such as issue trackers, databases or design files (`claude mcp add ...`; see Lesson 5.2). - **Skills** (Lesson 5.3) package procedures Claude loads when relevant, stored in `.claude/skills/`. - **Subagents** handle focused sub-tasks (for example, a test-writing agent) with their own context. - **Hooks** run your own scripts at defined moments (for example, formatting code after every edit). - **Plugins** bundle skills, MCP servers, subagents and hooks for easy sharing across a team. For developers building their own agents, the **Claude Agent SDK** packages the Claude Code harness as a library you can run on your own infrastructure. ## Worked example: an agency builds a client tool A two-developer agency in Dubai needs a small internal tool that pulls campaign data from a CSV export and produces a weekly PDF summary. With Claude Code, the lead developer writes a CLAUDE.md, asks for a plan, approves it, and lets Claude build the script in small steps with tests. She reviews every diff, adds a hook that runs the linter, and keeps the API key in an environment variable. A job that would have taken three days took one, and the code went through the same review as human-written code. ## Governance questions before adopting agentic coding - What can it access (repositories, production systems, secrets)? - Which permission mode will we use, and who approves risky actions? - How are AI-written changes reviewed, tested and attributed? - Where do secrets live (never in prompts or committed files)? - On Team and Enterprise, which admin controls and usage analytics are enabled? ## Hands-on - **Non-engineers:** solve one spreadsheet or small automation problem with the safe snippet pattern; test on a copy. - **Builders:** install Claude Code on a non-critical repo, run `/init`, edit CLAUDE.md, and complete one small task plan-first with tests. ## Pitfalls - Running unreviewed scripts on live data. - Giving an agent broad access to production. - Accepting large diffs you do not understand. - Letting CLAUDE.md go stale. ## How to measure success Tasks ship faster with no increase in defects, every AI-written change is reviewed and tested, and no secret ever appears in a prompt, log or commit. ## Video lecture: Coding help and Claude Code, for non-engineers and builders Lecture coming soon · 14 chapters · about 9 minutes. Read the full transcript below. 1. Claude and code 2. Why this matters for everyone 3. Two modes 4. The safe snippet pattern 5. Claude Code: first moves 6. A good CLAUDE.md 7. Extending Claude Code 8. Simple example 9. Worked example: a three-day job in one 10. Governance first 11. Try this now 12. Watch me do it, part 1 13. Watch me do it, part 2 14. Recap and next step ## Lecture transcript ### Claude and code You do not need to be a developer to get real value from Claude's coding skills, and if you are a developer, Claude Code changes how you work. In this lecture, non engineers will learn a safe pattern for formulas and small scripts, and builders will see how Claude Code works inside a real codebase, plus the governance questions every team should ask first. ### Why this matters for everyone Why should non engineers care about code? Because a surprising amount of office work is small, repetitive processing. Renaming files, cleaning a spreadsheet, moving rows between sheets. A ten line script can remove hours of that every month. But code also acts on real data, and a mistake can overwrite something important. Think of it like a power tool. Enormously useful, and you still wear safety glasses. The safe habits in this lecture let you get the benefit without the damage. ### Two modes There are two modes. Coding help in chat means asking for a formula, a script or an explanation of an error, and Claude can even run code in a sandbox to test it. Claude Code is different. It is Anthropic's agentic coding tool. It works directly in your codebase, reading files, planning changes, editing across many files, running commands and tests and using git, asking permission according to the mode you choose. It runs in the terminal, in VS Code and JetBrains, in the desktop app, on the web, and in Slack. ### The safe snippet pattern For non engineers, use the safe snippet pattern. Say you are not a developer and name the tool, for example Google Sheets. Describe the goal in plain English. Then ask for three things, the formula or script, a plain English explanation of each part, and three test rows including an edge case. If a script changes files or data, always test on a copy of a small sample first, and keep backups. And never paste passwords or API keys into prompts or scripts. ### Claude Code: first moves For builders, install Claude Code from Anthropic's official documentation, open your project folder, and type claude. Run the init command to generate a CLAUDE dot M D file, which holds project guidance Claude reads at the start of every session, such as build commands, conventions and what not to touch. Then ask for a plan first. Read the codebase and propose a plan to add a size guide to product pages, do not edit yet. Approve the plan, then work in small, testable steps. ### A good CLAUDE.md A good CLAUDE dot M D is short and specific. The stack. The commands to run lint and tests before saying a task is done. Folders it must never edit, like production config. Local requirements, for example that Arabic right to left and English layouts must both be checked. And a rule to keep changes minimal and explain any new dependency. Keep it current, because a stale guidance file quietly produces stale behaviour. ### Extending Claude Code You can extend Claude Code. MCP servers connect it to tools like issue trackers, databases and design files. Skills package procedures that load only when relevant. Subagents handle focused sub tasks with their own context, such as writing tests. Hooks run your own scripts at set moments, like formatting code after every edit. And plugins bundle all of these for sharing across a team. If you are building your own agents, the Claude Agent SDK packages the same harness as a library for your own infrastructure. ### Simple example A simple example. In Google Sheets, you want column D to say overdue when the date in column C is before today and the status in column E is not paid. Ask Claude for the formula, a plain English explanation of each part, and three test rows including an edge case. You get the formula, an explanation you can actually follow, and test rows, one overdue, one paid, one due tomorrow. Paste them in, confirm the results, then apply the formula to your real sheet. That is the safe snippet pattern in action. ### Worked example: a three-day job in one A two developer agency in Dubai needed an internal tool that turns a campaign CSV export into a weekly PDF summary. The lead developer wrote a CLAUDE dot M D, asked for a plan, approved it, and let Claude Code build the script in small steps with tests. She reviewed every diff, added a hook that runs the linter automatically, and kept the API key in an environment variable. A three day job took one, and the code went through exactly the same review as human written code. ### Governance first Before adopting agentic coding across a team, answer five questions. What can it access, repositories, production systems, secrets? Which permission mode will you use and who approves risky actions? How are AI written changes reviewed, tested and attributed? Where do secrets live, and the answer must never be in prompts or committed files. And on business plans, which admin controls and usage analytics will you switch on? ### Try this now Try this now, and choose your lane. If you are not a developer, pick one spreadsheet or file problem that annoys you every week. Use the safe snippet pattern, ask for the formula or script, a plain English explanation and test rows, and try it on a copy of a small sample first. If you are a builder, open Claude Code in a non critical repository, run init, edit the generated CLAUDE dot M D so it contains your real commands and rules, and complete one small task plan first, reviewing the diff before accepting. Either way, write down one thing you checked before trusting the result. ### Watch me do it, part 1 Let me show both lanes. First, the non engineer lane. I paste the safe snippet prompt. I am not a developer, I use Google Sheets, show overdue in column D when the date in column C is before today and column E is not paid. Give me the formula, a plain English explanation, and three test rows. Claude returns an IF formula with AND and TODAY, and explains each part. I open a copy of the sheet, type the three test rows, one overdue, one paid, one due tomorrow, and the results are exactly right. Only then do I fill the formula down the real column. ### Watch me do it, part 2 Now the builder lane. In a small repository for a weekly PDF report tool, I type claude, then run init. It generates a CLAUDE dot M D, and I edit it to list the real commands, npm test and npm run lint, plus rules, never commit secrets and keep changes minimal. Then I ask for a plan first. Read the code and propose how to add a currency column to the PDF, do not edit yet. The plan has four steps, and I approve it. Claude makes the change and runs the tests. In the diff I notice it added a formatting library, so I ask it to use the existing helper instead. Tests pass, and I commit. ### Recap and next step Recap. Everyone can use the safe snippet pattern, ask for explanations and test rows, and always test on copies. Builders get the most from Claude Code with a clear CLAUDE dot M D, plan first, small testable tasks and reviewed diffs. And every team should answer the governance questions before scaling. Your next step: non engineers, solve one spreadsheet problem this week. Builders, run init on a non critical repo and ship one small task plan first. ## Key takeaways - Anyone can use Claude for formulas, scripts and explanations; ask for explanations and test rows, and always test on copies. - Claude Code is an agentic coding tool that reads, edits, runs and tests code in a codebase with permission controls, in terminal, IDE, desktop, web and Slack. - A clear CLAUDE.md, plan-first working, small testable tasks and reviewed diffs give the best results. - Decide access, permission modes, review rules and secret storage before adopting agentic coding. ## Try it Use the non-developer prompt pattern to solve one spreadsheet or small automation problem you have. Test on a copy and write down what you checked. - [Previous: Writing and editing workflows that sound like you](https://optimizeall.com/learn/mastering-claude/writing-and-editing-workflows) - [Next: Connectors and the Model Context Protocol (MCP)](https://optimizeall.com/learn/mastering-claude/connectors-and-mcp) - [All lessons of Mastering Claude (Anthropic)](https://optimizeall.com/learn/mastering-claude)