Your AI Code Companion is Your Worst Practices on Steroids

It's pretty clear, even to the average consumer who doesn't write code, that the world of AI has changed software development in a way few people could have predicted. I think every software developer at some point within the last 4 months has been asked if they are afraid for the future of their career.

When I first downloaded Cursor, the most useful part was the tab completions, which at the time felt like a significant productivity boost. Since then, tools like Claude Code have come around and completely changed the way I go about development.

But one thing has become abundantly clear as I've used agentic coding agents more and more: my weaknesses as a developer have become amplified.

A link to this GitHub issue (opens in new tab) circulated through my development team about a week ago, and while it does make me chuckle, it also points out a very real issue. Unless specifically asked, your coding agent isn't going to contradict what you have to say.

My ADHD brain and I absolutely suck at chunking tasks up into smaller bits and pieces. I get excited and carried away about the big picture idea that I have and just decide to get started. I open up my terminal, I type in claude and I ask it to do what I want.

I mean, it should just work, right? That's what these tools promise to do.

A minute and a half later, I check back in, and Claude has found a place he thinks needs improvement and gone down a rabbit hole, breaking a lot of things along the way. This is a move straight out of my ADHD playbook.

So, attempt number two. I press esc to brake the cycle of the runaway freight train, press shift+tab and begin dictating to my terminal. I tell Claude: "This is my goal. I need you to assist me with breaking up this task into manageable pieces. Try to foresee potential issues. Where would you start?"

Well, now I've got a list of tasks and a rough estimate of what it's going to take to complete these tasks. I now ask Claude to complete the first, which he does, and then promptly engages the parking brake.

Developers are going to interact with these tools and land face-to-face with the things that they don't do well. This is something that's intimidating and hard to approach for many people across many different areas of expertise. We're at a crossroads, and the good developers are going to come out by recognizing this and adapting and changing their ways to become better. The not-so-great developers are going to let the agent do the heavy lifting and not even bother to be curious or introspective about why they seem to encounter the same behaviors over and over again.