Want to understand Claude Code the way its creators see it? This new video interview (in the comment) is essential.

I was going to introduce Claude Code’s creators—Cat Wu, the product manager, and Boris Cherny, the engineer from Anthropic—and possibly share old video links about them. A recent video interview with both of them has been released on YouTube. I’ll share this one instead!

If you are going to be spending any time with Claude Code, you’ll benefit immensely from listening to them both talk.

🎥 Video Interview

I watched the interview just now and think it will be a perfect introduction to them both. They are both calm and collected, as if they clearly understand where everything is headed. Since the video was just released, Boris and Wu talk about all the latest features. It may feel a bit advanced at times, but you can safely ignore some parts and come back later when you’ve developed some familiarity.

You’ll understand the philosophies and thinking behind the features and how they, other employees, and their big customers use them.

It’s a very high signal video. Admittedly, I’m familiar with most of what they discuss, having seen every released video of them. However, I’ll highlight some key insights that caught my attention which may be useful for you to dig into.

👨‍💻 Building for Developers

They build the product for themselves—the developers—using Claude Code while building Claude Code.

🚀 Model improvement considerations

The plan mode is a form of scaffolding that helps developers collaborate with Claude and prepare a suitable plan before execution. This is a best practice that always leads to better results because Claude has better context. However, as the models improve, the boundary of what we can accomplish without this planning phase expands.

This raises the question: will they build features knowing they will eventually be subsumed by model improvements? The answer is surprising. They’ll build anything that improves the developer experience today. They actively look forward to that event when the model improves and the scaffolding is no longer needed.

🤖 Multiple Subagents

They actively use multiple agents. They use over 10 subagents for code reviews. Since these can produce false positives, they have additional agents specifically looking for them. Overall, this leads to more reliable reviews.

Another use case for multiple agents is when migrating code from one framework to another.

But this can lead some power users racking up $1000/month bills.

They’re transparent that it’s a premium product designed to work best with their top model, Sonnet 4.5, which outperforms the current Opus 4.

✨ Miscellaneous

Although it’s geared toward code and developers, it’s also extensively used for non-coding purposes like writing, researching, problem-solving, and legal document preparation.

If you want to understand not just what Claude Code does, but why it’s designed this way, this video is required viewing.