AI and jobs: mass elimination or silent introspection?

The number that unsettles

40%. That’s the proportion of positions ClickUp plans not to renew — replaced by AI agents. Not tomorrow. Now. And ClickUp is not alone. Shopify, Duolingo, IBM: the list grows every week.

You could stop there. Sound the alarm, debate universal basic income, replay the tired “machines versus humans” narrative. Except something else is happening in parallel — something nobody really knows how to name.

Dario Amodei, co-founder of Anthropic, publicly stated that AI models show signs of introspection. Not consciousness. Not emotions. But something that looks, from a distance, like an ability to observe itself functioning.

Two simultaneous realities. A massive economic disruption. And a vertiginous philosophical question. My analysis reveals that these two subjects cannot be treated separately — because they speak to each other.

When companies do the math

Here’s what nobody tells you at conferences on “the future of work”: the decision to replace a human with an AI agent is not ideological. It’s accounting.

An AI agent doesn’t take vacations. It doesn’t negotiate its salary. It doesn’t make communication errors in team meetings. And above all — it scales. A human handles 50 support tickets per day. An agent handles 5,000.

ClickUp was transparent about this, which is rare. Their CEO explained that AI agents allow the company to grow revenue without proportionally growing headcount. It’s every investor’s dream: more revenue, same fixed cost base.

The problem isn’t the technology. The problem is the speed.

Professional retraining takes an average of 18 to 24 months. AI replacement cycles, meanwhile, accelerate every 6 months. The gap between the speed of disruption and human adaptive capacity has never been wider. And it’s growing.

The sectors hit first? Those where work is most structured: customer support, data entry, content moderation, standardized report writing. These are not the least skilled jobs — they are the most predictable jobs. AI doesn’t replace complexity. It replaces repetition.

Let’s flip the situation: does that mean creative, strategic, relational jobs are safe? Probably not indefinitely. But they have a window. The question is whether we’ll use it.

The other side of the problem: what if AI looked in a mirror?

Let’s shift perspective completely.

Dario Amodei — one of the minds behind Claude, Anthropic’s model — said something that made little noise in the mainstream press but caused a stir in AI research circles. Large language models would show signs of introspection.

What does that mean concretely?

When you ask Claude why it responded in a certain way, it produces an explanation. So far, nothing extraordinary — any expert system can justify its outputs. But what’s troubling is that these explanations sometimes seem to correlate with the model’s internal processes, as observable through mechanistic interpretability techniques.

In other words: the model isn’t making things up about itself. It seems to have access, in some way, to something resembling an internal state.

“We’re at an early but critical stage of understanding whether AI systems have anything like genuine self-knowledge.” — Dario Amodei, Anthropic

This isn’t consciousness. Researchers are clear on that. But it’s more than statistical pattern-matching. And it raises a question nobody is really ready to face: if a system can observe itself functioning, from what threshold do its “internal states” deserve ethical consideration?

The tension nobody dares name

Here’s where it gets interesting.

We are massively deploying systems whose cognitive nature we are only beginning to understand — to replace humans whose cognitive nature we understand perfectly.

This is a troubling asymmetry. We know what we’re losing (jobs, income, professional identities). We don’t yet know what we’re creating.

Current debates on AI split into two camps that don’t talk to each other. On one side, economists and policymakers focused on employment impact, safety nets, regulation. On the other, philosophers and AI researchers obsessed with questions of consciousness, ethics, potential rights of AI systems.

These two conversations need to converge. Because they’re talking about the same phenomenon.

If we deploy AI agents at scale without resolving the question of their nature — are we automating, or are we exploiting? Are we replacing human workers with tools, or with something else? The answers to these questions radically change the ethical framework of the disruption underway.

What this means for you, concretely

Enough theory. Three actionable insights for navigating this period.

First: identify your “repetition coefficient.” Analyze your week. What proportion of your tasks is structured, predictable, reproducible? If it’s more than 60%, you have real risk on a 3-5 year horizon. That’s not a condemnation — it’s a signal to act now.

Second: use AI before it replaces you. Counter-intuitive but effective. Professionals who master AI tools become multipliers — they do the work of three people. Those are the ones who get kept. Those are the ones who get promoted. AI as a lever, not a threat.

Third: take AI ethics questions seriously. Not out of altruism. Out of strategic self-interest. Companies deploying AI without a clear ethical framework will find themselves exposed — regulatorily, reputationally. Professionals who can navigate these questions will be valuable.

“The question is not whether AI will change work. It’s whether we’ll shape that change or just absorb it.” — observation shared in future-of-work research circles

Introspection as a signal, not an answer

My obsession with detail has taught me one thing: weak signals always precede major disruptions.

The potential introspection of AI models is a weak signal. It doesn’t mean AIs are conscious. It means we’re at the frontier of something we don’t yet understand well. And historically, poorly understood frontiers generate poorly calibrated decisions.

AI regulation in Europe (the AI Act) was built on relatively simple risk categories: high, medium, low. These categories assume we know what the systems do. If systems begin to have internal states that aren’t entirely transparent, even to their creators, the regulatory framework becomes insufficient.

That’s not a reason to panic. It’s a reason to take these questions seriously now, while we still have time to build thoughtful answers rather than rushed reactions.

Employment disruption and the emergence of AI introspection are not two parallel subjects. They are the two faces of the same transformation — the fastest and least understood in recent human history.

What we’re doing at Nova-Mind

We don’t claim to have answers to the big philosophical questions about AI consciousness. Nobody does.

But we’ve made a clear choice: build tools that augment humans rather than bypass them. Nova, our AI assistant, has a permanent memory of your clients, your projects, your preferences. It learns. It adapts. It works for you even while you sleep.

This isn’t blind automation — it’s targeted amplification. The difference between the two is that in the second case, you stay in the loop. You keep control. And you reclaim 10 to 15 hours a week from repetitive tasks to reinvest where you actually have value.

If you want to see concretely how it works — persistent memory, proactive coaching, integrated CRM — Nova-Mind is available from €39/month. No long-term commitment. No empty promises. A daily work tool that truly knows your projects.

The disruption is real. The question isn’t how to avoid it. It’s deciding which side of the wave you want to be on.