Ask HN: What are your views on using AI in software engineering?

Claude Code for devops / iac right now as IC. LLM writes most code, I manage guard rails, style, success criteria etc. To be fair, writing code was never really the biggest part of the job, it was always figuring out what people need, what solutions make sense for what team in what context, talking to people, documenting things, testing etc. CC can be helpful for this stuff also, particularly exploring options and doc gen.

The tool works more or less perfectly for my relatively simple use cases. IAC codebases tend to be smaller so context management easy. It could be more capable of course but from my perspective there are no major cons. Typically I have 3 or 4 repos checked out into workspace and the tool is managing how to "stitch them together" / integrate. Having a local workflow is pretty key, one missing piece for me right now is being able to easily iterate on work that can only be tested by running a pipeline (without manually shuttling data between tools).

I don't think I actually deliver any faster because in enterprise you spend most time in wait states blocked by slow internal processes (ticket for X, approval for Y, meeting to discuss Z, 20 minute deploy pipelines). It would require major org transformation to shift this, seems unlikely to happen in the next few years.

I will say, my job feels easier and my cognitive load lighter. I don't need to "switch modes" as often between high and low level thinking. It "feels" like having a fairly capable junior (who nonetheless needs careful management and supervision) to handle the details.

Claude code (janky AF but the best right now). I use it for everything, almost literally. I don't even commit anymore, I have a command to do a commit preflight then inc version, commit with good messages, merge and push. Earlier today I ran out of disk space while doing nix builds and I used it to run the unix commands to see where on my filesystem I could reclaim space, filter out certain classes of things, then create a script I could use to run a clean-up. I give it MCPs and have it manage my google analytics, track social media, add editor's comments to my novel manuscript, the list goes on.

The good thing about it is that the plan is a hell of a deal, and it works well enough to not be annoying.

"ticket for X, approval for Y, meeting to discuss Z, 20 minute deploy pipelines" I guess eventually this will also get addressed with AI, many thanks sharing your insights.