Skip to main content

Coding & Software Development

All levels
What you'll learn
  • Know what AI coding is genuinely great at — explain, generate, refactor, debug, translate, review
  • Run the golden loop: context in, plan, generate, read, run — and loop failures back as fresh context
  • Reach for prompts that pull their weight instead of vague one-liners
  • Internalize the two hard rules: verify by running, and never paste secrets

Whether you're learning to code or shipping production software, AI changes the loop. The winners treat it as a fast, knowledgeable pair — and verify everything it produces.

What it's great at​

  • Explain unfamiliar code or errors in plain language.
  • Generate boilerplate, tests, and first drafts of functions.
  • Refactor for clarity, and debug by reasoning about a stack trace.
  • Translate between languages/frameworks.
  • Review a diff for bugs and smells.

For real codebases, do this in your repo with Claude Code, which can read files, run tests, and edit with your approval.

The golden loop​

  1. Give context — the relevant code, the error, what you expected vs. got. Vague in, vague out.
  2. Ask for a plan on non-trivial changes before edits (Plan Mode).
  3. Generate the change.
  4. Read it — understand before you accept. You own the code.
  5. Run it — tests/lint/build. Never trust "this works" without running it.

The step that separates good results from bad ones is the arrow back to the top: when a test fails, you don't patch blindly — you feed the failure back in as fresh context.

Prompts that pull their weight​

Explain code + spot the edge cases

Explain what this function does and any edge cases it mishandles: {code}

Generate tests

Write tests for {function}. Cover the happy path and the edge cases. {code}

Debug from a stack trace

This throws {error}. Here's the code and stack trace. Find the root cause and propose a minimal fix. {context}

Hard rules​

:::warning Verify, and protect your secrets

  • Run and review generated code — it can be subtly wrong or invent APIs that don't exist.
  • Never paste secrets/keys into a prompt (Privacy).
  • For agentic/automated coding, lock down permissions and read Securing Agents. :::

Check yourself

0/3
  1. In the golden loop, what most separates good AI-coding results from bad ones?
  2. Why read generated code before you accept it?
  3. Which of these should never go into a prompt?
Key takeaways
  • Treat AI as a fast, knowledgeable pair — then verify everything it produces by actually running it
  • Context in, quality out: give the code, the error, and expected-vs-actual, never a vague ask
  • Ask for a plan before non-trivial edits so you review the approach before any code changes
  • Read generated code before accepting — it can be subtly wrong or invent APIs that don't exist
  • Never paste secrets or keys into a prompt, and lock down permissions before letting an agent code on its own

Next​