AI can write the sentence but not carry its weight. What experience, judgement and originality mean for organisations when production becomes free.
An AI can write about what I felt during my mother's final months. The sentences will be well formed. The words will be the right words.
What it writes will not be it.
There is nothing underneath those sentences. The model learned the words. It did not learn the weight the words carry.
That gap is where the new definition of human value begins. What AI cannot produce is the experience of having suffered, loved and lost, and having made meaning out of it. Experience is original. Production is not.
This is not a sentimental observation. It has direct consequences for how organisations describe their value, hire their people and price their work.
For most of working life, distinguishing genuine from fake was a background task. Content was expensive to make, so most of what reached you had passed through some filter.
That filter has gone. Text, image, voice and video now cost close to nothing to produce at volume. The volume arriving at any professional has increased and the average quality of it looks better than it is.
The skill this demands is not detection software. It is judgement: knowing what a real argument feels like, what an honest number looks like, where a source should exist and does not.
This skill develops the slow way. It comes from having been wrong before, from having trusted something plausible and paid for it. Which is precisely why it cannot be generated.
The philosopher Byung-Chul Han argued in *The Burnout Society* that the modern subject is no longer disciplined by an external authority but exhausted by an internal one. Nobody has to force us to produce. We do it to ourselves, continuously, and call it achievement.
AI arrives into that condition rather than into a neutral one.
The promise was that machines would absorb the repetitive work and return time. What has often happened instead is that the capacity for output rose and the expectation rose with it. The team that produced four documents a week now produces twelve, and the pressure did not fall.
Han's point matters here because it names the trap. If AI is used only to increase throughput, it deepens the exhaustion it was supposed to relieve. The organisations getting value from it are the ones deciding what not to produce.
Erich Fromm drew a distinction in *To Have or To Be* between the mode of having, where the self is defined by what it accumulates, and the mode of being, where the self is defined by what it does and experiences.
Applied to an organisation, the distinction is unexpectedly practical.
A company in the mode of having describes itself by assets: how many people, how many products, how many tools deployed, how many models integrated. All of this can now be matched quickly by a competitor with a budget.
A company in the mode of being describes itself by what it can actually do: judgement it has developed, problems it has solved before, relationships it has kept through difficulty. This cannot be bought at speed and cannot be generated.
The question for a leadership team is which of these two lists their value proposition currently sits on.
Consider a specific scenario, and it is no longer hypothetical for most organisations.
A competitor, or an unrelated actor, produces content in your brand's tone. Same vocabulary, same structure, same visual language, at volume and at low cost. The imitation is good enough that a casual reader cannot tell.
What remains yours?
Not the tone; that was copied. Not the format; that was copied. What remains is the part that came from having done the work: the specific client situation you handled badly in 2019 and learned from, the number you know is wrong because you collected it yourself, the position you hold that would cost you money to abandon.
Brand defensibility has moved from the surface to the substrate. Anything that can be observed can be reproduced. Only what was lived cannot.
Three concrete implications for organisations.
Rewrite the value proposition around evidence rather than adjectives. Descriptions built on qualities (innovative, trusted, expert) are now trivially reproducible in text. Descriptions built on specifics (this is what we did, this is what it cost, this is what we would do differently) are not.
Hire for judgement, not fluency. Fluency with AI tools is becoming a commodity and will keep becoming cheaper. The ability to look at plausible output and identify what is wrong with it is rarer and harder to teach. In interviews, the second is testable: show a flawed but polished document and ask what is wrong with it.
Protect the conditions in which experience accumulates. Judgement comes from having owned outcomes. If junior colleagues only ever review generated drafts and never carry a piece of work end to end, the pipeline that produces senior judgement quietly closes. This is a five-year risk that looks like an efficiency gain in year one.
Experience is original. Production is not.
Everything in this piece follows from that sentence, and the sentence is worth holding onto because it makes decisions easier. When you are deciding what to automate, what to keep, what to pay for and what to develop in people, ask which side of it the thing sits on.
Planning this conversation for a Turkish audience? Read the Turkish version.
What to do this week:
1. Take your organisation's value proposition and mark every claim that a competitor could reproduce with a good prompt. Look at what is left.
2. Add one judgement question to your next interview: a polished document with a real flaw in it.
3. Check whether anyone junior in your team has carried a piece of work end to end this quarter.
→ Enquire about a future of work speaker for your event
---
Cassie Kozyrkov · Explaining the Generative AI Value Gap (YouTube)
The argument is narrower and more practical than that. It is that judgement, originality and defensibility come from lived experience, and lived experience cannot be generated. Whether a model has inner states is a separate question. What matters commercially is that anything observable can now be reproduced, so value has moved to what was lived rather than what is displayed.
Give the candidate a polished document containing a real flaw: a number that does not follow, a source that would not exist, a conclusion the evidence does not support. Ask what is wrong with it. Fluency with tools shows up in a portfolio; judgement shows up in what someone refuses to accept.
Tone of voice is no longer a moat, because tone is reproducible at volume. Specific and verifiable claims are. Move the value proposition from adjectives towards evidence: what you did, for whom, at what cost, and what you learned. That material cannot be generated by someone who was not there.
No. The argument is about what to use it for. Using AI to raise output volume tends to raise expectations with it and leaves the pressure unchanged. Using it to remove low-judgement work and protect time for high-judgement work changes the outcome. The decision is about where the freed hour goes.
If junior colleagues only review AI-generated drafts and never own a piece of work end to end, the experience that produces senior judgement stops accumulating. The cost does not appear in year one, when it looks like efficiency. It appears later, when the organisation needs people who can tell that a plausible document is wrong.
After Dinner Speakers
Contents
