How I use a 20 year old engineering paradigm to keep my AI tokens usage optimized.

Not to create any suspense, the methodology I am talking about is Agile.

When I first started out as an engineer right out of college, I was the impatient bloke on the team. Always wanting to get things done fast, I hated writing/reading/updating Jiras(Issues or whatever it is called in 10 other such tools). Over time, as my responsibilities grew and projects got bigger, I finally understood the importance of tracking tools and the art of breaking down a system piece by piece, and I became the most elaborate Jira writer there ever exists on a team (Wherever I went, I created templates for issues and pushed to use them extensively, and the results almost always surprised everyone. Well, except for the impatient bloke on the team (not me this time, lol)).

So, when I got my hands dirty with Claude Code, I thought of developing projects in the same Agile manner, and it worked like magic. With the very first prompt, I let Claude run free like a madman and exhausted my standard 20$ plan’s hour window limit with just 2 prompts in total. Then I mustered my inner software engineer who had seen it all at 4 different global corporates. I entered plan mode and asked Claude to divide each project into phases, and each phase into Jiras/Issues - the smallest possible unit achievable independently- ranked them in order of priority, and then let Claude execute.

The result - I experimented faster. I caught bugs early on. I intervened in misdirections and now have to oversee each unit of development; I slowed down and built intentionally. As a result, my 20$ token limit suddenly started looking bigger. I saved a ton of tokens and was actually able to achieve workable units when I ran out of tokens, as opposed to getting stuck with nothing being built and waiting for token resets.

Though this is about Agile and breaking down software into smaller, iterative pieces, it triggers a larger thought in me - No matter what new, upcoming, emerging technology shows face tomorrow, the fundamentals and classic paradigms are not going anywhere. And even if they do, we can always learn a thing or two from history :)