How Claude Code's Interview Mode Changed My Workflow

· 5 min read ·
·
AI Claude Code Developer Tools Productivity Workflow

It’s the second day of 2026, and I stumbled upon something that fundamentally changed how I approach building features with Claude Code.

I was watching a YouTube video about “Interview Mode” - a concept that seemed almost too simple to be revolutionary. But after trying it, I’m convinced this is how I should be working with AI coding assistants.

The Problem I Didn’t Know I Had

Like many developers using AI coding tools, I’d fallen into a pattern: write a prompt, answer a couple of quick questions, then watch Claude spin up a bunch of code. Sometimes it worked perfectly. More often, I’d realize halfway through that it built something I didn’t actually want.

Take authentication, for example. I’d say “add authentication to my app” and Claude might implement JWT tokens. But what I actually wanted was a managed service like Clerk or WorkOS. By the time I realized this, Claude had already scaffolded an entire JWT system. I’d spend more time undoing work than I would have spent planning upfront.

The “Interview First” Approach

The concept comes from Tariq, who works at Anthropic. He tweeted about using a “spec-based” workflow:

“My favorite way to use Claude Code to build large features is spec-based. Start with a minimal spec or prompt and ask Claude to interview you using the Ask User tool. Then make a new session to execute these specs.”

The workflow is simple but powerful: Interview first, spec second, code last.

Instead of me prompting Claude with everything upfront, Claude prompts me. Instead of Claude guessing what I want, it asks for clarification. Instead of discovering buried assumptions during code review, I confront all those decisions when they’re cheap to change.

How It Works

The magic happens through Claude Code’s “AskUserQuestion” tool. You’ve probably seen this before - it’s the feature that pops up with multiple-choice questions mid-session. But the key insight is to use it proactively, not reactively.

Here’s the prompt I’ve been using:

---
name: interview
description: Interview me about the plan
argument-hint: [plan]
---

Read this plan file $1 and interview me in detail using the AskUserQuestion tool about literally anything: technical implementation, UI & UX, concerns, tradeoffs, etc. but make sure the questions are not obvious.

Be very in-depth and continue interviewing me continually until it’s complete, then write the spec to the file.

For big features or new projects, Claude might ask me 40+ questions. And I end up with a much more detailed spec that I feel I had a lot of control over.

My First Experience

I decided to test this with a relatively broad prompt: “Build me a Next.js app. Interview me about what I want.”

What happened next surprised me. Instead of Claude making assumptions and jumping into code, it started asking thoughtful questions:

  • “What am I actually building? A web app, a marketing site?”
  • “Who is the target audience?”
  • “What’s the primary goal of this application?”
  • “Do you need authentication? If so, what type - managed service or custom implementation?”
  • “What’s the expected traffic scale?”
  • “What’s your preferred styling approach?”

Each answer led to more nuanced questions. By the time we finished, I had a comprehensive spec that captured decisions I didn’t even know I needed to make.

Why This Works

The traditional prompt engineering approach was about crafting the perfect instructions in one go. But with spec-driven development, AI helps you discover what you actually want - because it might not be as clear as you think right away.

When you give Claude a prompt, Claude is going to make a ton of different assumptions. The interview process surfaces those assumptions early, when they’re cheap to change. Instead of reworking code later, you’re making decisions upfront.

You’re narrowing the space of all possible solutions that Claude could generate. Instead of discovering buried assumptions during code review, you confront all those decisions when they’re cheap to change - before Claude writes code and uses all those tokens to generate something you don’t want.

The “Slow Down to Speed Up” Philosophy

Most people’s first experience with AI coding tools is: put in a prompt, answer a couple questions, then watch it build. But the idea with interview mode is really to slow down to speed up.

The more time you spend planning, the less time you’re going to spend reworking what Claude might have already done. It’s counterintuitive - you’re using an AI tool to go faster, but you’re intentionally slowing down the initial process. Yet the results speak for themselves.

Interview Mode vs. Plan Mode

Claude Code has had Plan Mode for a while, which is excellent. It explores your codebase, designs implementation files, and writes a plan. Interview mode isn’t meant to substitute Plan Mode - it can be used as a precursor.

The workflow becomes:

  1. Interview - Claude asks you questions to understand requirements
  2. Spec - Write a detailed specification based on the interview
  3. Plan - Use Plan Mode to design the implementation
  4. Code - Execute the plan

This sequence ideally leads to much better outcomes because you’ve thought through the requirements before planning, and planned before coding.

Setting Up Interview Mode

You can use the exact prompt Tariq shared, or build your own variation. The key is to trigger the AskUserQuestion tool and let Claude guide the conversation.

I’ve also added this as a skill in Claude Code, so it triggers automatically when I mention certain keywords. The nice thing with skills is that Claude will follow the process and latch onto those native tools within Claude Code.

The Realization

I don’t understand why this concept of Requirements → Specification → Design → Code is treated as a new insight. It’s how engineering has been done correctly for hundreds of years.

But I think the hype around “vibe coding” - where you just prompt and get code - glossed over those specific activities.

We assumed our coding assistant was some kind of psychic that “knows” what you want. But treating AI like a team member - discussing project goals, having them explain algorithms, discussing standards and styles, then moving forward on coding - that’s the right approach.

To do otherwise is like throwing mud against the wall and seeing what kind of pottery you get.

My New Workflow

Now, whenever I have a large feature to build, I don’t try to get it all within one prompt. I actually try out the interview process. I ask Claude to interview me about the plan, leveraging that AskUserQuestion tool.

The results have been surprising. I feel more in control. I catch edge cases earlier. And I spend less time undoing work that was built on wrong assumptions.

Try It Yourself

Next time you have a large feature to build, I encourage you to try this out. Don’t just try and get it all within one prompt. Actually try out the interview process. Ask Claude to interview you about your plan, leveraging the AskUserQuestion tool.

You might be surprised at the results you get.

Have you tried Interview Mode in Claude Code? I’d love to hear about your experience. Reach out on LinkedIn.