In a world of Perplexity, Dia, Google, Brave, and countless other advanced search tools, I kept coming back to this decision. I must have revisited it around 50 times over the past couple of weeks, and I ultimately ended up spending a few hundreds on it and paid for a search product, Kagi

I realised somewhere between "making it easy to consume information" and "urge to make decisions", the art of discovery is fading away.

Couple of months ago, we began building first-class plugins for Claude, Codex, Copilot, and Cursor. Each of these tools was still inew, changing almost every day. Almost every AI plug ended up breaking or slowing me down, even when it was just an embedded search across their own documentation. 

Regardless of the problem, what actually moved me forward was browsing the docs myself, almost like flipping through a newspaper:spotting the bits that were relevant to us, and starting from a scratch‑level hypothesis before asking another machine for help.

I resisted an AI search engine to synthesise information before I’d formed a hypothesis of my own. Strongly.

Few months later, we find someone reviewing Scalekit and call this out https://lnkd.in/gVB8h-mM