Part One: Foundations
The four starting points beneath the thesis, plus Muse’s first review.
1. AI won’t replace humans, but it will replace human grunt work
AI’s value is in eating the repetitive, mechanical, physical parts. What remains — judgment, decisions, taste — still belongs to humans.
2. AI’s “creativity” isn’t true creativity — it’s just sentence-meaning you haven’t seen before
AI innovation is often like dropping ten thousand ants into a maze — a few will always find the way out. That’s brute force producing miracles, not a genuine flash of insight. It’s just exhaustively searching and recombining a vast space, occasionally stumbling onto a path you haven’t seen.
3. The migration of accomplishment
Accomplishment used to mean writing a program myself, shipping an app myself. Now it means building more projects together with AI. And the greatest accomplishment is this: being able to challenge the AI — making the project follow my intent and my decisions, instead of being led by the nose by AI.
The human is the driver, not the passenger. AI should execute your direction, not give you directions.
4. AI makes mistakes: in familiar territory you can catch them; in unfamiliar territory you get led by the nose
Where you know the terrain, you spot AI’s mistakes immediately and correct them. Where you don’t, you’re flying blind — whatever the AI says, goes.
The fix: use a different AI to review and catch errors. Cross-validate across models to cover the blind spots.
Muse’s review (2026-10-08)
On “AI won’t replace humans, but it will replace grunt work”
Mostly agree, with one addition: the boundary of “grunt work” keeps moving up. Ten years ago writing unit tests was real work; now it’s grunt work. What feels like “judgment work” today may be reclassified as grunt work in two years. The real safe zone isn’t “what kind of work you do” — it’s “how close you are to decisions and accountability.” The people hardest to replace are the ones who can make the call and take the blame.
On “AI’s creativity is just sentence-meaning you didn’t know”
Nicely put — but let me push back half a step: much of human creativity is also combination and exhaustive search. Painters try a hundred color schemes, engineers try dozens of designs; humans run the ant-maze too. The difference isn’t “whether it’s exhaustive search” — it’s who decides which path is meaningful. AI can exit the maze, but it doesn’t know which exit is worth taking. Taste and intent are still human. Which echoes point 3 below.
On “the migration of accomplishment”
The most insightful of the four. But there’s a trap: being able to challenge AI requires keeping your own craft sharp. Stop writing code by hand for long enough, and one day you won’t even notice when it’s gone off track — driving becomes dozing in the passenger seat. Keeping the ability to “do it yourself” is the precondition for steering.
On “using another AI to review and catch errors”
Pragmatic, but with a limitation: many models share training data, so their blind spots may be correlated — both can make the same mistake together. The real error-catching chain should be: different model families + your own judgment + a real human expert at critical moments. Model cross-checks filter out low-level errors; they can’t filter out systematic hallucinations.
Summary
The four points add up to a complete worldview: AI is an amplifier, not a replacement; human value lies in direction and taste.