The feeling I get after working with agents like Claude Code or Cursor is that suddenly,

anything and everything I want to do is at my fingertips—and I just want to do it.

I don’t think too much;

I just get an idea, spawn a terminal,

explain my concept in detail to the clanker,

and off it goes.

For example,

On a weekend, I was exploring some new thing like I usually do.

I wanted to experiment with a new code editor called Helix,

So I downloaded it, used it a bit and tweaked the settings to my liking.

I was missing a Copilot LSP for my Helix editor

And nobody had ever made one apparently,

so I asked the clanker to implement it for me.

It got an MVP done in a couple of minutes.

I was actually able to use the Helix Copilot LSP,

and it gave me autosuggestions while I was typing.

Looked great, however

After using it, I noticed all these small details I wanted to fix

so I asked the clanker to handle those too

another thing

oh but this implementation did something off

and suddenly, I looked at the clock and it was 3 AM.

Yes

I had implemented a Copilot LSP for Helix

but it wasn’t practical to use

and I didn’t really care that much,

I was just testing Helix for an afternoon,

to see if a simpler text editor would work better with agentic development

(it didn’t, neovim ftw btw)

But I realize now: I just got attacked by the AI Vampire!

AI coding agents like Claude Code mess with dopamine levels,

It’s the “slot machine” feeling they create.

Every time I send a prompt, it’s like pulling the lever on a terminal:

I feel a rush of anticipation, wondering if this next generation will be the “big win” that solves the problem perfectly.

Even when it fails, the near-miss makes me want to go “one more round” to fix it.

This cycle of instant, unpredictable rewards keeps me glued to the screen for hours without realizing it.

While fixing one issue, another idea pops while I’m waiting for it to finish.

Therefore, I have to implement strategies that work best for me so I don’t get sucked in too much.

The problem is not actually building things, the feeling I get after writing a lot of code I thought about hard, is good usually. If I am implementing things by hand, my plan gets better and better. I am able to reflect more during the process, and I get a dopamine hit only after a long time of effort into a problem solving topic. The problem is getting sucked into “pulling the AI lever”

I think it varies from person to person how “hooked” someone can get when interacting with these kinds of systems.

In my case, since I am quite sensitive,

I realize not everyone is like this, but I am very easily sucked into things like video games, social media, and now AI agents.

I don’t play video games,

I’ve run away from social media,

Not because I don’t like them, but because I know I can get sucked in too much.

Now that at work there is a system that can have the same effect on me,

I realized I have to be careful.

So I am trying to have some rules.

I never use “Agentic Development” at home.

I usually work from the office, and the physical separation helps me to keep a separation from my home activities.

If I can’t go to the office (maybe a doctor appointment or something), then I go to a local cafe if I want to do agentic development.

Having the “lever” at home after a day of “pulling a lever” is not good (yes, even with spec driven development it can be that if the specs are simple)

Having a physical separation so I don’t get the itch to do something while relaxing at home is important.

The AI Vampire is some name that already existed and I thought this sentence should apply:

“The AI Vampire will only take over your life, if you invite him to your house”