What AI is doing to soft skills
And here's the surprise. You might think: well at least empathy, communication, leadership are safe.
Not quite.
AI can write emails that sound empathetic. Generate feedback that's constructive and supportive. Run a coaching conversation - and do it quite well, if the person on the other side doesn't know they're talking to a model. Create narratives, build persuasion, write speeches.
So where is the real boundary?
This is exactly the core of Amodei's thesis. Not in what AI can simulate. In what requires being a human with history, with context, with understanding of why people do what they do.
Amodei's thesis in plain terms
Soft skills aren't data resources. They're the capacity for interpretation.
Take an example. AI can analyze a transcript of a client conversation and identify that the client was speaking slowly, using negatively loaded words, and asking defensive questions. That's data.
But to understand why this specific person, in this specific organization, with this specific history in the industry, behaves this particular way - you need something more. Historical context. An understanding of motivation. The ability to make sense of behaviors through the lens of experiences the model knows nothing about, because it has never lived through anything.
Humanities - literature, history, philosophy, sociology - don't teach facts. They teach how to ask questions about meaning, motivation, and consequences. And that is genuinely difficult to teach a model.
Not because models are stupid. Because understanding humans isn't a data extraction task. It's an interpretation task.
And interpretation - as Paul Ricoeur wrote - is always someone's interpretation. Always embodied in a particular perspective, history, presence. Ricoeur was a twentieth-century philosopher and one of the key theorists of hermeneutics. He argued that understanding a text - and more broadly, the world - is not data extraction but a dialogue between the reader and what is being read. Nobody understands "objectively." They always understand from somewhere.
A model has no perspective. It only has words about perspectives.
Implications for leaders and organizations
Amodei says AI primarily amplifies and increases productivity - it doesn't replace. She observes "a mixed picture." And she argues that in a world with capable AI, what will matter most is: communication, empathy, collaboration, critical thinking.
This isn't a lone voice. Similar observations appear among education researchers and labor economists - the argument that automation has historically eliminated tasks, not occupations, and that skills difficult to codify become more valuable. There's no single study that "settles" this - but the direction Amodei points to is consistent with a broader conversation in the field.
What does this mean in practice for the leaders I work with?
A few observations - no verdicts, just questions:
Communication as a profession. Not pretty emails. Understanding who is listening, what they're hearing, what they actually need. AI can help with the words. It can't read the room.
Critical thinking - but what kind? Not "be able to calculate without a calculator." Be able to ask: is this the right problem? Are we measuring the right thing? Does this number mean anything? AI delivers answers faster than ever before. The cost of asking the wrong question is rising.
Collaboration under ambiguity. Projects where the answer is known are increasingly being solved by AI. What remains are the ones where the answer is contested, where values collide, where you need to negotiate meaning. That's human work.
History and context. A leader who understands where their organization came from - what its traumas, successes, and founding myths were - understands why certain changes don't take hold. AI can read the report. It didn't live through the last five years inside that company.
My five-year bet? I'd wager on interpretive, narrative, and relational skills. Not instead of technical ones - alongside them. A person who understands both AI and people is rare today and will become more valuable.
How do you build both at the same time? Not by alternating between a prompts course and a literature course. Through practice in places where the two meet - projects where technology has to answer a human problem. Where a well-functioning API isn't enough; you need to understand why someone has the problem at all and what drives them. That kind of work teaches both things at once.
A question to close with
Daniela Amodei runs a company building some of the most advanced language models in the world. And when asked about the skills that matter most - she talks about empathy, critical thinking, and literature.
What have you done lately to develop something AI won't replace?
Because if the only answer is "I'm learning prompts" - you might be optimizing for the tool instead of the value the tool is supposed to amplify.