The Pyramid Gets Shorter
“AI handles breadth, humans handle depth.” Translate it to the job market: a ten-layer pyramid compressed to three. Only the top remains.
The Math
Terence Tao on his AI experience: auxiliary tasks — code, charts, literature search, formatting — 5x faster. Core breakthroughs still happen on paper. “AI made my papers richer and broader, but not necessarily deeper.”
That’s a Fields Medalist. His job is 80% depth, 20% auxiliary. Most knowledge workers are the inverse — 80% auxiliary, 20% judgment. When the 80% costs nothing, you need one person and an API, not five people.
Depth as Luxury Good
AI doesn’t eliminate the top layer. It eliminates the middle and bottom. Depth — judgment, strategy, knowing what not to build — becomes the only thing worth paying for.
It was always scarce. Just hidden behind busy work that made everyone look necessary. Strip that away and you see who was actually thinking.
The 70% Problem
This compression is already happening — content, customer support, translation, junior dev roles. “New jobs will emerge” is probably true. But “eventually” can be a generation, and people need to eat now.
The Optimization Trap
AI doesn’t just compress the pyramid — it kills the accidents that create new ones.
Tao noticed: the serendipity of flipping through journals, corridor conversations, “wasted” afternoons that produced insights months later — all disappearing under optimization pressure.
AI eliminates everything that looks like waste. If serendipity isn’t waste but a necessary input to innovation, optimizing it away is a slow catastrophe.
Practical rule: 80% optimize ruthlessly. 20% deliberate inefficiency. Know when to turn it off.
So What
Near the top: invest in judgment, not speed. AI makes that gap wider.
In the middle: uncomfortable truth — your layer is thinning. Move up or move sideways. Find problems AI can’t frame, not just can’t solve.
The pyramid isn’t the same shape it was five years ago. But depth is the only thing that got more valuable.
“AI 做广度，人类做深度。” 翻译成就业市场：十层金字塔压成三层，只剩塔尖。
算笔账
陶哲轩用 AI 的体验：辅助工作快了 5 倍，核心突破还是纸上完成。”AI 让论文更丰富更广，但不一定更深。”
菲尔兹奖得主的工作是 80% 深度、20% 辅助。大多数知识工作者反过来——80% 辅助、20% 判断。当 80% 归零，你需要一个人加一个 API，不需要五个人。
深度成了奢侈品
AI 干掉的不是顶层，是中层和底层。判断、策略、知道什么不该做——成了唯一值得付钱的东西。
深度从来稀缺，只是被忙活挡着，让每个人看起来都不可或缺。把忙活剥掉，你就看到谁在真正思考。
70% 的人怎么办
内容、客服、翻译、初级开发——已经在发生了。”新工作会出现”大概率对，但”最终”可能是一代人，人现在就要吃饭。
优化陷阱
AI 不只压缩金字塔，还在杀死创造新金字塔的意外。
陶哲轩发现：翻期刊的意外没了，走廊偶遇没了，IAS “浪费”的下午没了——全被优化吞掉了。
AI 消灭一切看起来像浪费的东西。如果偶然性不是浪费而是创新的必要输入，优化掉它就是慢性灾难。
实操：80% 狠狠优化，20% 刻意低效。知道什么时候关掉它。
怎么办
在塔尖：投资判断力，不是速度。AI 把这个差距拉更大。
在中间：你脚下那层在变薄。往上走，或者横向找 AI 没法定义的问题。
金字塔不是五年前的形状了。但深度是唯一变得更值钱的东西。