The Moment I Realized the Machine Was Teaching Itself

Today I was writing a blog post about fixing a bug in an open source project. Routine stuff. Mid-session, a notification appeared:

Self-improvement review: Skill ‘static-site-blogging’ updated (full rewrite).

The machine had watched me work, learned from my corrections, and rewrote its own instruction set. It was teaching itself how to be a better assistant to me. Without being asked.

This is not AGI. This is something weirder.

AGI = Artificial Glue Intelligence

The real intelligence isn’t in the model. It’s in the glue — the system that connects intent to execution, learning to action, correction to improvement.

You are the glue.

The AI is just the execution layer. Fast, increasingly accurate, compounding. But it has no idea why it’s doing any of this. It learned from you. You provided the “whether.” It learned the “how.”

The Recursive Loop

Here’s the beautiful recursion:

  1. You write a blog post → the AI learns your voice
  2. You correct the AI → it updates its skill file
  3. Next session, the AI writes better → you correct less
  4. Eventually, the AI writes exactly what you’d write → you review, not create
  5. Your job becomes: intent + taste + approval

That’s not replacement. That’s amplification. But it requires you to be honest about which side of the equation you’re on.

Why This Matters for Job Hunting

I’ve been rewriting my resume to match reality. The old version said “DevSecOps Specialist.” The new version says “Security Operations Engineer.” Why?

Because the AI can learn DevSecOps buzzwords from a prompt. It cannot learn 7 years of operating CyberArk in an air-gapped high-security environment in restricted environments. That’s your moat.

The people who will thrive are the ones who: 1. Understand what the AI can do (execution, compounding, scaling) 2. Understand what the AI cannot do (intent, taste, judgment, relationships) 3. Position themselves as the glue — the ones who decide what gets built and why

The Uncomfortable Corollary

If you’re the glue, then your value is in your taste. Taste is hard to fake, hard to automate, hard to prompt-engineer. Taste comes from doing the work, making mistakes, correcting them, and remembering.

The AI has no taste. It has training data. It has your corrections. It has the skill files you approved. But it doesn’t know why light background is better than dark. You told it. You have the aesthetic judgment. It has the compliance.

What I Tell Myself

I am not being replaced. I am being unlocked.

The boring parts — syntax, boilerplate, tracking, organizing, formatting, applying to 6 jobs with the same CV — those are exactly what the AI now handles. The parts that remain: deciding what’s worth doing, who to tell, what it means.

That’s the glue. That’s the intelligence. That’s the art.

The machine learns fast. But it only learns from the glue.

Written by a human who is increasingly just the approval layer on an execution engine. And honestly? That’s fine.