AI == Junior Dev who doesn't learn

I recently came across the talk by Vincent Quigley regarding how he uses Claude Code.

Here are my notes from his talk where he explains how he lets AI write the majority of his initial code (80%).

And why treating AI like a junior dev who doesn’t learn has become his mental model these days.

The Coding Pivots

In his career, he has had 4 major pivots in the way that he codes.

One interesting thing is that each transformation happened faster than the previous way of doing things.

How he develops with AI

I use AI mostly “to think with” as I’m working with it towards the code that ends up in production.

Expecting perfection on attempt one is like expecting a junior developer to nail a complex feature without context.

The Context Problem + Solution

AI can’t retain the learning that it has done between sessions. We have to manually provide it.

And here is how he connects other tools with Claude Code

PR Process

Challenges

Cost

Well, I’m seeing this pattern across multiple orgs. And yeah, Claude Code is quite expensive if you use usage-based billing.

It looks like he has 8% of the company spend on Claude Code 😅

Thoughts

Background Agents could be a next way of doing things

What does not work?

AI doesn’t learn from mistakes. You fix the same misunderstandings repeatedly. Your solution: better documentation and more explicit instructions. AI confidently writes broken code claiming that it’s great. Always verify The context limit problem. Large codebases overwhelm AI context windows. Break problems into smaller chunks and provide focused context.

Happy building-with AI!