Are we all builders now?

A few years ago I was sitting at work using the newest AI tool (GitHub Copilot) that gave me the ability to "tab" complete a line. An LLM sat in the shadows predicting what probably would come next and often times it was wrong, but often times it was right. It felt like magic at that point because just a few characters into typing and the rest of the line might be automatically completed with a tab. I evolved into this engineer that could tab complete a line and then correct it slightly. In that timeline it felt like I was given a superpower that didn't break the bank.

Now I can watch a non-technical person dispatch an agent that follows our code style & guidelines and produces a pull request that isn't that bad. Sure it needs some work as long as humans are still maintaining codebases, but it's crazy to think a few years prior this was unheard of.

I remember this argument early on in my career of "build vs buy" which basically means does a new client coming in the door have a solution available they can buy and configure or do they really need a custom built piece of software. Sometimes that's not the easiest research as you have to jump on sales calls and dig into a 3rd party piece of software before discounting it. All of this just being one of the things you'd like to do before you start writing code - a little discovery session.

Now, as we countdown the months till 2027 the landscape of clients and requests have changed dramatically. No one has time for a lengthy discovery phase or research when you build a working solution with Claude in a few hours. This extends into my open source development just as strongly. You are getting bug reports & pull requests with an enormous amount of text that try and break down every little detail possible.

I'm really torn with all of this, because at one point that much context/research provided by another user showed they put some serious effort into understanding the problem & providing a solution. You'd look at that issue/pull request and think "dang - this person did some work. I appreciate it" and reciprocate the effort in reviewing it. These days you'll have an influx of pull requests with confident robust descriptions and the only person that has to spend some human brain cycles is the maintainers.

You could toss in the towel and just let an AI manage your codebase, which is borrowing the production term "lights out manufacturing". The digital version of this is exactly what it sounds like - an LLM is the judge and jury when it comes to merges, reviews and more. There are a few of these approaching "lights out" I've been watching and the throughput is insane. There is no chance any human fully understands anything in the codebase anymore, but does any human need to at that point?

I'm not yet at that level of comfort, because at the end of day if a system is catastrophically collapsing I'm probably getting involved at work. If I release a buggy build of fastlane or Apktool in the open source space - I'm getting pinged. For that reason if I have to be the one to solve a problem I can't yet give up the ownership of the codebase to a machine yet.