I’ve been spending some time building on Claude Code. I started with a simple product: a morning briefing for myself.
The goal was simple. I don’t want to be checking news sites. I don’t want news briefings or notifications. I don’t want any scrolling. I don’t want to be stuck in doom cycles.
Instead, I wanted the most important news from my industry with clear “so what’s” as they pertained to me and my job. I wanted some global awareness news that didn’t require a lot of attention but was enough to not catch me off guard. And some local news.
And I wanted all of this in a one-stop source to create a high-quality information diet.
It started with a simple spec. But as time passed, the spec became more and more complex.
And as simple as this briefing was, it took about two days of iteration before getting it to a place where I was happy.
While some of this was time spent understanding how to get the best out of Claude Code, a big part of this was improving the spec.
I took away 3 lessons –
(1) If you’re able to take the time to plan and think through what you want, you’ll go much faster than if you just start building. Go slow to go fast.
(2) It helps to be systematic when we approach iteration. Initially, I used to just ask Claude what happened and make a fix directly. Every such fix then broke three other things.
Things got better when I go into the rhythm of compiling the execution errors, and asking Claude Code how I could improve the spec to avoid it.
(3) Finally, you can get to 80% quickly. But the final 20% takes a long time.
It takes time to do things well. Do fewer things. Do them well.
A Learning a Day 