No Complaints

This happened early on, back when I was still getting a feel for what the agentic development process could actually do.

I'd been poking around the code that Claude had generated, and I'd added Swagger to the API so I could see the endpoints being called and inspect what was coming back. As I was reviewing things, it hit me: I hadn't defined any commenting standards. There was no guidance in my engineering specs about how the code should be commented — what level of detail, what style, where comments should go. The codebase was readable, but it was largely uncommented.

So Claude and I had a conversation about it.

And here's the thing that still makes me chuckle. It gave me three options — essentially small, medium, and large. Light touch with just the critical stuff, a balanced approach with method-level and section-level comments, or comprehensive documentation-grade commenting throughout. We picked medium.

Then I said, "Great. Go ahead and run through the entire codebase and add the comments."

And it did.

Just — did it. The whole codebase. No pushback, no negotiating, no passive-aggressive compliance where half the comments say // does the thing. It went file by file, understood the code, and added meaningful comments throughout.

Now think about that for a second. Imagine you're leading a development team, and you decide mid-project — after a significant chunk of the code is already written — that you want to retroactively add comments to everything. Every file, every service, the whole thing. You know exactly how that conversation goes. At best, you get eye rolls and a week of grudging compliance. At worst, you've got people updating their resumes.

Claude didn't care. It's not that it was obedient — it's that the task was genuinely useful and it just executed. No ego, no context about how things were done before, no "but we've already shipped three milestones without comments." Just: here are your three options, you picked this one, done.

There's a broader point here for anyone experimenting with agentic development. If you've got an existing codebase and you want to dip your toe in, this is a nearly risk-free way to start. Point Claude at your code with your commenting guidelines — or let it propose some — and let it do the tedious work. You get immediate, visible value, and you start building confidence in what the system can do before you hand it anything more consequential.

And if you later decide to change your commenting standard? Same thing. Update the spec, tell it to go again. No complaints.

← Back to Blog