an LLM-assisted post

I was probably eight or nine when my mum put me into art tutorials. My sketching skills today stem comes from those classes. Thank you, Ma. And thank you, Miss Priya.

Theres one thing I remember very clearly; I hated it when Miss Priya corrected my drawings by sketching over them. I wanted to figure it out myself. No hints. No shortcuts. I want to get it right by myself, evne if the sketched did genuinely require corrections.

That mindset never really went away.

I've carried it straight into engineering. I love getting close to the bare metal. I like knowing how things work underneath. That includes LLMs. Funny, considering how much I've talked about AI and productivity.

But lately… I've been giving in.

I've started using LLMs for the boring stuff—boilerplate code, repetitive glue, even this blog! And honestly, it's insanely fun watching it turn rough business logic into actual working code in seconds. With a bit of care, it feels inevitable that LLMs become the layer most people talk to when building data applications.

Over the last two weeks, I've shipped more than I normally would in months. Things that used to take five times longer just… don't anymore. And getting things done matters.

Having said that, the better part of about this productivity boost is that I'm still diving deep when I want to. The same LLM I use to move fast also helps me understand fundamentals—AI papers, aerodynamic designs, random curiosities I'd usually take to forums and wait days for.

So yeah, I'm giving in.

Not because I stopped caring about doing things “right,” but because I realized I can move fast and still learn.

And yes—I'm probably very late to the party.