Programming with Claude Code is boring.

Programming with Claude Code is boring. I prompt, wait, review, prompt, wait, repeat. I don’t get into a flow and end up scrolling some feed. I put up with this because it was the shiny new thing and increased my output. But if the work becomes babysitting this prediction machine, I’m not going to last. Don’t get me wrong, I don’t miss writing code. I miss the problem solving. I miss the problem finding. I miss the exploration. I miss the refining. I miss the insights that would make everything fall in place. I think there is a way to recapture this fun, but I don’t know what it is yet.

My current workflow with Claude Code Or any of the others is to use the research, plan, implement flow I first heard about in Advanced Context Engineering for Coding Agents. This produces markdown files I need to read, review, and edit Really, I ask the machine to edit .

I use a skill Fancy name for reusable prompt to research. I wait. Once the research doc is generated, I review it. Maybe another prompt to have the machine look into some place it missed or reconsider its findings.

I prompt for the plan. I wait. I review the plan doc. Usually there are things that need fixing and fine-tuning. I prompt for that. I wait. Repeat until done.

I prompt for the implementation. I wait. I review the code Lately, the machine does its own reviews and I review the review . Prompt for changes. Repeat until done.

My mind wanders while I wait, and it always wanders into some feed.

I used to explore. I would browse the codebase. Imagine what could be. Maybe write it down, maybe not. If that didn’t point me in the direction of a solution, I might spike out a solution, play in the REPL. Once I had a rough shape and direction, I would think through what I needed to ensure the problem was solved. I would start with a test that drove me in the direction of the solution. As I went, things would be figured out. Holes in my thinking would be discovered. I might run into an unexpected issue. I would have to figure out a way around that. I would get insights for better names for things, a better design, maybe even a completely different approach. I miss this.

Now the prediction machine gives me options and I choose one.

It’s easy to miss the holes with the prediction machine. It generates coherent-sounding output, not true output. I have to be vigilant not to confuse the two. And since my mind wanders while waiting, I’m not that vigilant.

Time does not move backwards. The prediction machines are here to stay. They are truthy rather than true, and that’s good enough for producing code. Most problems in software development do not call for exploration and creative design. The prediction machines are good at those. I don’t want to go back to writing code because the code was just the output of a process. So that’s the new challenge: how do I work with these prediction machines to do more of what I enjoy about programming?