There are two popular ways to build with AI right now: Option A: Throw a top-tier frontier model at an unoptimized prompt, watch your API bill melt your CFO's soul, and call it 'innovation'. Option B: Act like a software architect.
Sit down, design the context, route tasks methodically, and nurture a system of specialized agents. Brute force intelligence works right up until you check under the hood. Methodical engineering, when done well, simply scales right..
Is it perfect?? Absolutely not!
If to be honest, neither if I was to hand code it. Settling for 'second best' is a choice now, simply because building it right doesn't take forever anymore. You can actually aim for perfection without burning down your roadmap.
I used to have a very predictable relationship with technical debt.
In the past, whenever an elegant, high-craft solution presented itself, reality would strike: 'That will take too long or half of the team's sanity.'
Today, perfectionism isn't just optional, it is within our reach!
That custom infrastructure or niche utility tool that used to trigger a never ending debate? It’s now a productive weekend with my an agent. A months build became a very focused week.