Post contents
Onboarding in the Ever-Changing World of AI
I’ve recently joined a new company, and I wanted to share my experience from my first week- what onboarding looked like in a world that’s changing fast thanks to AI.
The company (and the scale shock)
The company I joined is very popular in the AI and open-source world, over 30k GitHub stars. It’s growing in popularity, it’s taking names, and it’s moving rapidly. It still feels like a startup, it's all about moving fast, delivering value and making the customers happy without too much overhead and bureaucracy. They have a lot of products spanning a wide horizontal space: 50+ example apps and integrations spread across 3 huge monorepos, around 20 open-source packages, and growing. By no means is the codebase a joke.
Day 1: the part that will never change
The last time I was onboarding was 3 years ago, and the initial process hasn’t changed much:
- Spend your first day going through boring docs
- Learn about procedures and workflows
- Get familiar with the tech, the assistants, and the company goals
The first day in any company has been written in stone and probably won’t change even if we reach AGI.
Day 2: where it gets interesting
That being said, the second day is where it gets really interesting. I was always fast at adapting. If you’ve watched anime and JJK, I’m like Mahoraga. If not, you’ll leave this article confused as to what I mean.
Normally, I’d start being productive in the first 3-5 days, depending on the project. But this was one of the most complex codebases I’ve encountered, and yet I was productive on day two.
I went into the codebase, saw quick wins, and I got to keep what I kill.
In the first 3 days, I added things like pkg.pr.new, migrated the monorepo from Jest to Vitest, and a couple of other improvements. That was pretty mind-blowing. And of course, it was heavily AI-assisted.
Here’s the weird part: I had no real intimate knowledge of the codebase. If you asked me where some functionality lived or how to change a specific thing, I couldn’t tell you. And yet the value was still real, just the Jest → Vitest migration sped up the testing part of CI by ~35%. I still haven’t fully caught up, but AI lets me bring value despite that, which is kind of amazing.
Reality: Productivity isn’t the same as understanding
But let’s have a real talk here. Even though I’m much more productive much sooner, I’m still completely useless at adding business logic into the codebase without actually understanding the system. I got a task to implement a feature across two monorepos, and I had a lot of trouble with it. The first part (in the first monorepo) was easy: the Linear ticket was very detailed, I could fully explain to Claude what it needed to do, and once it did, I iterated over it and got it working rather quickly.
The issues came after.
It was time to move into the neighboring repo and carry the changes over. The other monorepo is much more complex, and I ran into problems that AI can’t really solve for you:
- Missing .env setup Cryptic build issues due to using Windows (yes I know, Windows and I both suck for this)
- Five intertwined packages I don’t fully understand yet
I implemented everything the way I thought it should work… and it didn’t.
I used AI to try to debug it, and it didn’t help too much. In the end, I went back to the good ol’ reliable method: console logging everything. I’m still deep in it, keep your fingers crossed for me.
So what am I trying to say with all of this?
Onboarding has changed a lot!!
You can move faster, you can “fly” sooner, and you can deliver early wins without full context. But without deep understanding, you still can’t go far. Onboarding looks easier, but if you don’t go through the motions, if you don’t really learn the architecture, the workflows, and the shape of the system, you’re not going to contribute significant, impactful changes. I keep saying that knowledge and understanding of architecture, codebases, and workflows is more crucial than ever, and this week just proved it for me again.
AI helps me implement things faster, especially universal improvements like devtool usage, CI speedups, and general maintenance. But to bring true value to a company, the human is still as important as ever.
We ain’t going nowhere… at least for the next 3 months until a new model drops and we get replaced on LinkedIn again. Keep being awesome and keep delivering value. Improve yourself, and everything will grow around you as a side effect.