Asking Better Follow-Up Questions

We touched on the conversational nature of Claude earlier, but mastering the "follow-up" is where you go from being an AI novice to a power user.

Most people get a bad answer, sigh, delete the prompt, and try again from scratch. That wastes time and tokens. Instead, you should steer the conversation.

The "Pivot"

Sometimes Claude goes down a path you don't like. Maybe it generated a React component using class components, and you want functional components with Hooks.

You don't need to rewrite your original prompt. Just pivot.

"That works, but our codebase exclusively uses functional components with React Hooks. Please rewrite this without using any Classes."

Claude remembers the entire context of the component it just wrote, so it will instantly refactor it.

The "Zoom In"

If Claude gives you a massive architectural plan, you might be overwhelmed. Don't ask for the whole plan again. Zoom in on the specific part you care about.

"The overall architecture looks solid, but I don't fully understand step 3 (the Redis caching layer). Zoom in on that. Give me a detailed breakdown of how the data actually flows into Redis and write the specific Node.js code for that connection."

This keeps the overall context intact while forcing Claude to provide microscopic detail on one specific pain point.

The "Critique"

This is perhaps the most powerful follow-up technique. You can actually ask Claude to review its own work.

"Look at the code you just generated. If you were a senior developer reviewing this PR, what security vulnerabilities or performance bottlenecks would you point out? List them, and then provide the updated code fixing those issues."

It is genuinely shocking how often Claude will spot bugs in its own code that it missed on the first pass, simply because you forced it to change its perspective to "critic" instead of "creator."

Mastering these follow-ups turns a rigid prompt-and-response cycle into a fluid, collaborative engineering session.