Skip to main content

Coding & Software Development

All levels

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 what this function does and any edge cases it mishandles: {code}
Write tests for {function}. Cover the happy path and the edge cases. {code}
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. :::

Next​