The skills framework was built for a world where humans do the execution.
Hard skills: technical. Writing code, reading a balance sheet, modeling risk. Soft skills: interpersonal. Communication, leadership, judgment under pressure. It's a clean taxonomy, and for most of the twentieth century it held.
AI breaks it at the execution layer. It writes code. It reads documents. It models faster than any analyst. The hard skills most people spent years acquiring are now table stakes — not because they stopped mattering, but because the machine can cover them.
What remains isn't a soft skill. It's something the old taxonomy didn't name, because humans had no choice but to execute themselves.
The new differentiator is knowing what to ask. Whether the output is right. What problem is actually worth solving. When to override. When to trust. These aren't soft skills — they're not interpersonal, they're not about managing people or reading a room. They're something else: a capacity to orient before the machine starts moving.
I built my entire research and publishing system with Claude Code. I'm not a developer. I built it because I could frame the problem clearly, evaluate the output honestly, and iterate from failure. The machine executed. I supplied the judgment.
That's the category. Not hard, not soft. Call it orientation: knowing what good looks like before the machine starts.
People are racing to learn skills the machine already possesses. The more useful question is what you build that it doesn't.
Adapted from Asymmetrical Collage.
https://www.asymmetricalcollage.com/articles/2026-04-11-what-remains/
Hard skills: technical. Writing code, reading a balance sheet, modeling risk. Soft skills: interpersonal. Communication, leadership, judgment under pressure. It's a clean taxonomy, and for most of the twentieth century it held.
AI breaks it at the execution layer. It writes code. It reads documents. It models faster than any analyst. The hard skills most people spent years acquiring are now table stakes — not because they stopped mattering, but because the machine can cover them.
What remains isn't a soft skill. It's something the old taxonomy didn't name, because humans had no choice but to execute themselves.
The new differentiator is knowing what to ask. Whether the output is right. What problem is actually worth solving. When to override. When to trust. These aren't soft skills — they're not interpersonal, they're not about managing people or reading a room. They're something else: a capacity to orient before the machine starts moving.
I built my entire research and publishing system with Claude Code. I'm not a developer. I built it because I could frame the problem clearly, evaluate the output honestly, and iterate from failure. The machine executed. I supplied the judgment.
That's the category. Not hard, not soft. Call it orientation: knowing what good looks like before the machine starts.
People are racing to learn skills the machine already possesses. The more useful question is what you build that it doesn't.
Adapted from Asymmetrical Collage.
https://www.asymmetricalcollage.com/articles/2026-04-11-what-remains/