The Two-Step Pattern That Keeps Me From Going in Circles with Claude Code

I've been using Claude Code since it was released over a year ago. I use planning mode most of the time for anything beyond simple changes. But I've found a specific two-step refinement that works even better.

Occasionally I'd skip it and end up in a frustrating loop. Claude builds the wrong thing, I try to correct it, and it drifts further. Every single time, the problem was me not being clear enough upfront.

This two-step pattern keeps me honest. It's not complicated. But it catches miscommunication before it turns into wasted time.

Step one: ask Claude to switch into planning mode and explain back what you're asking for. Not to solve it. Just to tell you what it thinks you want. This matters because what we think we're saying is not always what we're actually saying. Especially if you're speaking into a microphone instead of typing. You say something, Claude interprets it, and those two things can be surprisingly far apart.

Step two: once you both agree on what the problem is, ask for a plan. Read the plan. Make sure it makes sense. Then build. By the time you start writing code, Claude understands the problem well and you've agreed on the approach. Miscommunication caught at this stage costs you a minute. Caught after Claude has built the wrong thing, it costs you hours.

You don't always need this. If you're changing a few lines of code, just ask. Skip the ceremony. But for anything with medium complexity, this pattern pays for itself every time. For really complex tasks, it's the same pattern but phased. Break the work into chunks and align on each one. A bit more work, but the same idea.

The real question is: when can you skip it? There are no hard rules for this yet. Sometimes I skip the pattern and it works fine. Sometimes I skip it and get burned. My recommendation: always do the two steps until you build up the muscle memory. Over time, you start recognizing what you've done before, what's similar enough to try in one shot. That instinct comes from experience, not from a checklist.

The pattern always works. It just takes a bit longer. That's the trade-off. And I can tell you from plenty of scars that a few extra minutes of alignment beats hours of debugging a miscommunication that happened at the start.

If you're using Claude Code or any AI coding tool, try this for a week. Align first, plan second, build third. See if you stop going in circles. I'd bet you will.