Kilo Code

I wished I would have started with Kilo Code instead of Windsurf. It's such a cool platform and toolset for trying out LLMs. One place and you get API access to pretty much every major model out there all at the prices you'd pay for making and setting up all the accounts on the providers themselves.

And you're not just limited to their offerings either, unlike Windsurf and others that locks you into their ecosystem through subscriptions if you don't want to use Kilocode's cool API system then just setup your other providers they offer a lot out of the box and are always adding more. For example you can setup Gemini CLI or Qwen Code to get access to their free tiers while getting to work in Kilo Codes interface and with all their added features like options for code indexing, MPC server market place and ...etc. If you're bringing your own API key from someplace like Groq, Google...etc. You don't evne need a Kilo Code account to use Kilo Code.

Thus far in my AI/LLM journey Kilo Code and Cherry Studio AI are my two best finds. Kilo Code for all things coding, and Cherry Studio AI for all things Chat.

I tried LLM-assisted coding with Cline and Kilo Code, and it has been a terrible experience. LLMs have the memory of a goldfish and the attention span of a squirrel on stimulants—they forget not only what you told them to do, but also what they themselves were doing literally a second ago. At this point, instead of writing elaborate prompts asking them to generate a function that matches a specification, you might as well tell them to write anything at all—the end result would be about the same.

So I started experimenting with agent modes, MCPs, skills, agent definitions, and project rules. It didn't make much difference. What really shocked me was when an LLM acting as an agent in Kilo Code couldn't even tell me what rules it was supposed to follow while working. This was despite all the agent definitions, documentation files, and everything else I had carefully put together. So why did I spend all that time writing a dozen Markdown files if they don't actually do anything?

Everything related to LLMs feels like one giant scam. Just like the NFT craze, where we were told that half the economy would soon revolve around selling .jpg images.

Now we're being told that LLMs are useful for just about anything beyond translation. Yeah, right... That's why users are fed up with AI slop, and programmers—whose productivity was supposedly going to skyrocket thanks to AI—often prefer... not to use AI at all.

As for me, the only thing LLMs have been useful for is wasting my time.