Picking up the pace a bit with AI

I've been using Claude Web and Claude Code in tandem for quite a while now with good results. Currently, my process looks like this:

Plan out a piece of work on paper.
Break it down into small chunks, ensuring they have a logical order.
Fire up Claude Code (CC) in VS Code.
Ensure the staging area is clear.
Activate plan mode and iterate on a plan until it's looking good.
Allow CC to edit automatically.
See where it lands, read code, tweak if needed.
Commit.
Repeat steps 4-8 until the work is complete.

There generally hasn't been a lot of tweaking necessary with this approach. I think this is down to a few things:

Spending lots of time understanding the task at hand before jumping in - if you don't know what you want, you're going to muddy the waters of Lake Context.
Feeding tasks to CC one at a time so it doesn't get confused.
Setting CC up for success - @ing relevant files or prior art, explaining the motivations for the change, explaining the goal, pasting screenshots if relevant, and only then asking for what you want.

The caveat here is that I'm getting CC to work on an existing codebase. There are established patterns, a set of components/hooks/utils, and an explanation of what the project is and what it does. While the tasks CC works on are more than just glueing these pieces together, it does seem that this particular type of scenario is one that allows for a consistent solid output.

I'm still trying to balance writing my own handmade ✨artisanal✨ TypeScript with "getting Claude to do it". That said, I have softened up a bit and I'm definitely warming up to the idea of finding a happy medium between pressing all the keys myself and becoming a manager for a team of AI agents.

In short - plan mode, detailed instructions in a logical order, and incremental commits are getting me pretty far at this point. I've been delivering significant pieces of work with this approach.

I've also been exploring skills so I can equip CC with both specific capabilities at project-level and personal preferences at user-level, but I've got more testing to do on this front. It feels like there is a lot of value here but I'm trying to avoid creating a really bloated setup that has lots of bells and whistles just for the sake of it. Making the conscious choice to grow my setup along with my understanding of the tooling has been working well so far.