In this hard time, I know that everyone wants instant results. With prompts, they want their needs satisfied like magic. Abracadabra! You have a product.

Nowadays, everyone is talking about AI. On Threads, a lot of programmers brag about their vibe-coding agents. They build end-to-end systems using nothing but prompts and markdown files to tell the AI which rules to follow. I’m not anti-vibe-coding, since I’ve tried it too, building things with agents and MCP. But in the end, I feel it’s not my cup of tea.

What I prefer is using AI only when I really need it, for example, to answer questions I’d otherwise have to search for on Stack Overflow, or in cases where my code linter, application logs, or system logs give no clues.

I code in Neovim, and there’s a plugin called CodeCompanion. You have to check out their website, it’s an awesome plugin! It has agentic features, MCP support, and other nice tools, but honestly, I never use those anymore, except for the editor context: #buffer, #buffers, #diagnostics.

What I do with CodeCompanion is as simple as asking what, why, and how, not generating whole chunks of code that could burn through all your tokens. I usually only need $6 of Claude API credit per month. I don’t know about you, but that fits my needs just fine.

The worst outcome of using AI without control is that it sometimes hallucinates answers based on old documentation, which may give us a deprecated response. And if we don’t know the programming language fundamentals, it will lead us to error-prone code and untraced security holes. Please use AI wisely, as your companion.