From Roasting LinkedIn to Rethinking How I Build Software (Context > Vibes)

A few weeks ago, I built LinkedinScope as a weekend experiment — the idea was simple: roast LinkedIn profiles with AI, for fun, a bit of noise, and some subtle social commentary on the over-optimized personal branding epidemic.

It blew up more than expected.

But beyond the virality, it did something to me as a builder.

I realized I was no longer coding by “vibe”.

Before this, I was deep into Karpathy-style vibe coding — prompt the LLM, get something cool, ship fast, tweak later. But building LinkedinScope, with a very specific tone, behavior, and sequence, made me realize something: if you want AI to deliver exactly what you need, vibes aren’t enough.

You need context engineering.

Roasting someone’s profile sounds easy, until you need:

• consistent tone (witty, but not cruel),

• custom logic based on job roles,

• multilingual output with nuance,

• and fast turnaround without hallucinations.

It pushed me to create structured flows, dynamic system messages, isolated context blocks, and fallback layers. I wasn’t just prompting llms, I was crafting its information environment.

The joke tool forced me into a more rigorous way of building.

I still love the speed and magic of vibe coding. But I’ve now seen how far context engineering can take things — especially when product behavior needs to be repeatable, fine-tuned, and scaleable.

Funny how building a roast generator helped me grow up as a dev.

Let me know if others here feel this shift too — from vibes to systems.

And if you’re curious: https://linkedinscope.com