jinko-niwashi

Your "don't fucking touch that file" experience is the exact pattern I kept hitting. After 400+ sessions of full-time pair programming with Claude, I stopped trying to fix it with prompt instructions and started treating it as a permissions problem.

The model drifts because nothing structurally prevents it from drifting. Telling it "don't touch X" is negotiating behavior with a probabilistic system — it works until it doesn't. What actually worked: separating the workflow into phases where certain actions literally aren't available. Design phase? Read and propose only. Implementation phase? Edit, but only files in scope.

Your security example is even more telling — the model folding under minimal pushback isn't a knowledge gap, it's a sycophancy gradient. No amount of system prompting fixes that. You need the workflow to not ask the model for a judgment call it can't be trusted to hold.

The instincts can absolutely be developed faster with AI — if you set it up right. I work with an AI partner daily and one thing I've noticed is that it's a brutal mirror: it exposes gaps in your thinking immediately because it does exactly what you tell it, not what you meant.

That feedback loop, hundreds of times a day, compresses years of learning into months. The catch is you need guardrails — tests that fail when the AI drifts, review cycles you can't skip, architecture constraints it must respect.

That's what builds the instincts: not the AI doing the work for you, but the AI showing you where your understanding breaks down, fast enough that you actually learn from it. Just-In-Time Learning.

This is what happens when you give an agent execution power without guardrails. The tool isn't the problem — the absence of governance is. In my setup I treat the AI as a junior dev with root access: every destructive operation requires explicit human approval, and the session context includes hard constraints on what it can and can't touch.

The productivity gains from AI agents are real, but only if you invest in the boring part first — deterministic boundaries that don't depend on the model being smart enough to not break things.