A staff engineer's journey with Claude Code

To all the engineers using claude code: how do you submit your (well, claude’s) to review? Say, you have a big feature/epic to implement. Typically (pre-ai) times you would split it in chunks and submit each chunk as PR to be reviewed. You don’t want to submit dozens of file changes because nobody would review it. Now with llms, one can easily explain the whole feature to the machine and they would output the whole code just fine. What do you do? You divide it manually for review submission? One chunk after another?

It’s way easier to let the agent code the whole thing if your prompt is good enough than to give instructions bit by bit only because your colleagues cannot review a PR with 50 file changes.

I regularly write big MRs, then cut them into 5+ (sometimes 10+) smaller MRs. What does Claude Code change here?

I split my diffs like I've always did so they can be reviewed by a human (or even an AI which won't understand 50 file changes).

The 50 file changes is most likely unsafe to deploy and unmaintainable.

Do whatever you want. Tell it to make different patches in chunks if you want. It'll do what you tell it to do.

The split seems artificial now. Before, an average engineer would produce code sequentially, chunk after chunk. Each chunk submitted only after the previous one was reviewed and approved. Today, one could submit the whole thing for review. Also, if machines can write it, why not let machines review it too? Seems weird not to do so.

Practically - you can commit it all after you're done and then tell it to tease apart the commit into multiple well documented logical steps.

"Ask the LLM" is a good enough solution to an absurd number of situations. Being open to questioning your approach - or even asking the LLM (with the right context) to question your approach has been valuable in my experience.

But from a more general POV, its something we'll have to spend the next decade figuring out. 'Agile'/scrum & friends is a sort of industry-wide standard approach, and all of that should be rethought - once a bit of the dust settles.

We're so early in the change that I haven't even seen anybody get it wrong, let alone right.

I built a tool to split up a single PR into multiple nice commits: https://github.com/edverma/git-smart-squash

Will the LLM take responsibility for the bugs and bad code introduced by the review? If it does and I'm free, then go for it.

Not sure I follow. The limitation has never been about the developer being able to write a complex feature in one MR. It has always been about the other developer not being able to review a complex MR.

So far, nothing I've seen convinces me that machines can (yet) write or review code autonomously (although they can certainly be useful as assistants). Maybe some day.