🇻🇳 Also available in Vietnamese: AI không làm mình nhanh hơn, nó thay đổi công việc của mình

A year ago, whenever I was designing a new tool, a good chunk of my energy went into “how are we going to build this.” Today, that question has almost disappeared from my day-to-day.

I can focus almost entirely on what I expect from the product: the intended behavior, the edge cases, the business constraints. Implementation no longer dominates my attention — from an engineer who writes code, I’m becoming more of a designer of what I build.

But that doesn’t mean less work. It means different work. A good chunk of the time saved on dev goes straight into verification: making sure the deliverable actually meets the request, that it respects security constraints, internal standards, team conventions. That phase can’t be delegated, and it takes time.

And in a company, a tool can’t just stop at an MVP. It has to stay maintainable and operable over time. That doesn’t mean reading every line of code, but it does mean keeping a clear picture of the implementation choices and the overall logic. Without that, I’d be building up debt I’d only discover months later.

Concretely, it plays out on two fronts at once:

  • On new work: I can test the technical viability of an idea in a few hours instead of several days, and go from concept to a first usable deliverable very quickly.
  • On existing systems: I can finally tackle what had been sitting in the backlog for months — rewriting tools that had become obsolete, never quite priority enough to get the time.

Both matter equally. But if there’s one area where I see a gain I couldn’t really measure before, it’s technical debt: it’s no longer just “I’ll move faster on the next project,” it’s “I can finally deal with what I never had time for.”