Ai on MoskitoHerohttps://moskitohero.com/tags/ai/Recent content in Ai on MoskitoHeroHugoen-usTue, 12 May 2026 22:10:00 +0000Opening the Mind of a Thinking Agenthttps://moskitohero.com/post/2026-05-12-opening-the-mind-of-a-thinking-agent/Tue, 12 May 2026 22:10:00 +0000https://moskitohero.com/post/2026-05-12-opening-the-mind-of-a-thinking-agent/Reflecting on my experience of using AI agents for coding, I start wondering what is really hiding behind the concept of a "thinking agent". Time to take a look inside its brain…​

My brain is a dumpster tonight.

I kind of felt lazy all day, so I decided to rely on agents to leverage most of the work I had to do. The result is that I now feel totally depressed and tired. Everything was wrong with that day’s work.

It was tempting, at first. I thought I would have AI do the boring bits, type less on my keyboard and become an orchestrator. I thought it would give me more control over my priorities and help me dash through the day like a warrior, like a knight proudly riding his horse.

Now it’s 6 p.m. It feels like the horse has been trampling on my back all afternoon and is now chewing on my hair.

What did I get from today’s coding session? I remember giving instructions to the agent. I remember waiting for it to reply. I remember being unsatisfied with what it proposed as a solution. I remember guiding him step by step through the process. I remember looking at the screen in disbelief as it took five minutes to move a piece of code from one file to another.

I could have done it in seconds. Still, I was kind of relieved not to do the boring chores.

Looking back, I feel that what really happened is that I switched off my brain and surrendered it to the Machine. Now I feel ashamed.

All that time the machine was thinking. I wasn’t.

Thinking…​

That word intrigued me. So I clicked on it. A toggle opened. And man, was that ugly…​

What AI companies advertise as thinking is actually a computer program mumbling for ages, balancing ideas and their opposites, trying to figure out what exactly I meant, taking U-turns, battling against itself on details so insignificant it would drive anyone with a sane mind totally nuts. This is not thinking. This is vomiting words by the gallon. This is a brute-force parody of a french political debate.

My experience of thinking is not filling up pages and pages of words based on the likelihood of their occuring in a single sequence, and then taking these pages of writing to write yet another round of pages of nonsense until I decide I am confident enough to tell you I have finished thinking.

My experience of thinking actually implies growing up as a person and as a professional, backed by an intimate network of neurons I have patiently connected throughout the years by the means of experience, reading, learning, meeting, confronting ideas.

My experience of thinking is tied to the dose of dopamine I get when, Eureka! I finally found the solution. I didn’t get my fix today. I gave it up to Claude, who took five long minutes of water-heating, climate-warming and deforesting to spit out an unconvincing solution while I looked bluntly at my screen like a Half-Life zombie crying out for brain.

🧟‍♂️🧟‍♂️🧟‍♂️ Braaaaaiiiin... 🧠🧠🧠

Luckily enough, I got woken up by a message telling me I had reached my daily quota. That’s when I realized the pain and frustration I had been inflicting myself.

Thinking is a pleasure. I will not let a machine with no feelings take it away from me.

Let’s wrap it up.

I am not anti-AI. That would be stupid. Like being anti-electricity, or anti-dishwasher, or anti-railway. I just want to keep my agent saddled-up tight underneath me like a faithful horse I can control. I want to use AI, and never let it be the other way round.

I feel it is time to review my AI deontology. This is the plan:

  • Remove uneeded CLI agents I have installed just "to try it out".

  • Remove IDE plugins, extensions, apart from code completion stuff. That is often useful.

  • In Zed, check "disable AI"

  • Keep Claude Code (company subscription) installed — Claude desktop was already uninstalled a long time ago — and stick to these rules:

    • Only reviews and ideation

    • No thinking models

    • No write access to anything whatsoever. Yes, even elisp.

And for the rest, I will stick to good old Organic Intelligence.

]]>