Generative AI writes legal briefs, analyzes medical images, and produces production-ready code. The fear in the C-suite is understandable. But the data tells a different story.

Dario Amodei, CEO of Anthropic, put it succinctly in a recent conversation with podcaster Dwarkesh Patel: AI may write 90% of the code. But that doesn't mean we need 90% fewer engineers. Those are two entirely different things.

McKinsey's study "Agents, Robots, and Us" from late 2025 confirms: only a fraction of the skills base actually becomes obsolete. The World Economic Forum projects a net increase of 78 million jobs worldwide by 2030. 170 million new roles emerge, 92 million are displaced. And the Jevons Paradox is already playing out: demand for AI-fluent professionals has grown sevenfold in two years.

AI creates more work of a different kind, not less. The opportunity amounts to $2.9 trillion in annual value creation in the US alone. But only for companies that get the human factor right.

‍

The Jevons Paradox: Why Efficiency Creates More Work, Not Less

A 19th-century economic principle explains what's happening right now. William Stanley Jevons observed in 1865 that more efficient steam engines didn't consume less coal. They consumed more. Because falling costs per unit unleashed demand that had previously been uneconomical.

The pattern repeats. When ATMs were introduced, the number of bank tellers doubled: from 250,000 in 1970 to over 500,000 by the early 2000s. Spreadsheets grew the accounting profession by 75%. More efficient technology lowers the cost per unit and unleashes demand that was previously uneconomical.

For AI, this means: design your organization around the assumption that successful AI adoption creates more work of a different kind, not less.

‍

Five Imperatives for Decision-Makers

1. Define Your AI Strategy Before Deploying Tools

80% of all AI projects fail before reaching production. Usually not because of the technology, but because of missing strategy. Successful companies map use cases, data landscapes, and organizational readiness before selecting tools. This strategic groundwork is complex enough that external AI strategists and integration partners with cross-industry experience help avoid the most costly mistakes and keep the overall system in view.

2. Build AI Competence Systematically

Jobs requiring AI skills command a 56% wage premium over comparable positions. But self-directed learning isn't enough. It takes structured workshops and hands-on training, tailored to real workflows, real data, and concrete use cases. Experienced AI practitioners who design training embedded in daily work create a multiplier effect across the entire team.

3. Don't Underestimate the Need for Expertise

The more AI automates, the more demanding the remaining human tasks become. Lisanne Bainbridge's "Ironies of Automation" from 1983 rings truer than ever: automation removes the easy parts and makes the hard parts harder.

A concrete example: AI-powered legal research hallucinated in 17 to 33% of cases. Without domain expertise, these errors remain invisible. Even seemingly straightforward document structuring for intelligent retrieval fails without deep domain knowledge due to chunking errors, polysemy, and missing ontology modeling.

4. Respect the Contradictions

The prevailing narrative in many boardrooms: AI cuts costs by reducing headcount. The data says otherwise. The potential amounts to $2.9 trillion in annual value creation in the US alone, but only for companies that get the human factor right. Those who treat AI purely as a cost-cutting instrument forfeit most of the value.

5. Take the Transitions Seriously

42% of companies abandoned most of their AI initiatives in 2025. A dramatic spike from just 17% the previous year. Gartner predicted as early as 2024 that at least 30% of generative AI projects would be shelved after proof of concept. The leap from prototype to production is where most value is lost.

Experienced implementation partners, from data engineering and document intelligence to agent architectures and evaluation frameworks, are the difference between a failed investment and a system that actually works. This is even more true in environments with additional challenges such as the need for local models and proprietary data.

‍

The Real Point

The age of AI is not the age of human obsolescence. It is the age of human elevation, but only for companies that invest in their people's ability to work with AI rather than betting on AI instead of people.

The real question isn't whether AI replaces you. It's whether you learn to orchestrate it.

‍