AI only works in context

This piece first appeared in German. The English version is a rewrite rather than a line-by-line translation, and the German original stays online in the archive: Was Studien wirklich zeigen: KI wirkt nur im Kontext.

More than two-thirds of the 2,773 executives in Deloitte's generative AI survey, published in January 2025, expected that 30 percent or fewer of their experiments would be fully scaled within three to six months. The experiments exist, and the move into daily work mostly does not. That gap has little to do with the technology, and a great deal to do with the frame it is put into.

Few technologies have spread this fast, and few years have seen more experimentation, investment and publication. A pattern is emerging underneath all of it: AI produces value once it is embedded in human thinking, decision-making and organizational logic. Studies from 2024 and 2025 keep arriving at the same place. Technology alone creates no progress, and context, judgment and leadership are what turn possibility into effect.

Speed is not direction

At first sight AI looks like an accelerator, something that lets a team analyze faster, write faster and decide faster. Speed replaces no sense of direction.

BCG's AI at Work study found that only 13 percent of employees see AI agents deeply integrated into their daily workflows, and that only about half of frontline employees use AI tools regularly. Most people use the tools occasionally, with no systemic connection to anything. What is missing is the link between the technology and a purpose: who uses AI, when, in which decision context, and toward what end.

Without those questions it stays a gadget. With them it becomes an instrument for judgment.

Companies are measuring the wrong thing

Many organizations still define success through efficiency: time, output, cost reduction. AI reinforces exactly that logic, and reinforces its limits along with it.

The organizations that achieve an effect do three things differently. They define clear decision processes, and experienced staff get involved to check the results. AI is treated as part of a system, never as an end in itself.

Efficiency is good, and it should never replace the thinking that gives it direction.

Trust comes from understanding

The global study on trust in AI by KPMG and the University of Melbourne reports that only 46 percent of people are willing to trust these systems. More than half, 56 percent, say they have made mistakes in their work because of AI, and 66 percent rely on its output without checking whether it is accurate.

That is how quickly complexity turns into convenience, and following AI blindly costs exactly the thing that distinguishes us, which is critical thought. Judgment appears where technical results get understood, placed in context and, where necessary, corrected.

Every organization needs two things for that: transparency about how the systems work, and space for reflection. Understanding how a model arrives somewhere is the precondition for deciding whether it arrived in the right place, and there is no shortcut around it.

The second wave is a mental shift

The World Economic Forum describes a clear trend in its work on AI in industry. Most companies remain in pilot mode, and only a minority run AI inside scalable, governed structures.

What separates them is posture, and the question stops being what AI can do for us and becomes how AI changes the way we think, decide and lead.

That second wave is a mental transformation, and only secondarily a technological one. It asks leaders to interpret results, to own decisions, and to place the machine in the context of the business.

Experience is the corrective

Experience is what corrects AI, because experience means having been wrong before and having learned from it, which no amount of training data supplies.

The most successful companies maintain formal mechanisms for examining AI results critically before those results have operational effect, and those mechanisms rest on experience, skepticism and responsibility.

In consulting, what decides an engagement is the understanding of which data actually matters, and the speed of the model decides very little. Legal work turns on what the system might have missed more than on what it found in the contract. And no algorithm in investment banking replaces a feel for market psychology.

The more the machine takes over the doing, the more valuable the thinking becomes, and with it the judgment of experienced people.

Leadership is the translation layer

Technical results have to be translated into cultural, strategic and ethical context, and that translation is what leadership now consists of.

Leading in this period means producing meaning, and not controlling technology. Anyone who manages it connects two forces: the precision of the machine and the experience of the person.

Judgment can be institutionalized

Judgment is a competence that can be trained, shared and built into an organization. Very little about it is accidental.

Three things are worth doing. The first is space for reflection, in the form of regular reviews where results are questioned and put back into context. Support mentoring and shadowing, so younger people learn how senior decision-makers think and not only what they decide. And tie AI competence to responsibility: not everyone gets to deploy whatever a model produces, and responsibility stays human.

What grows out of that is a learning culture that AI strengthens.

AI is an occasion for thinking

Across all of these studies the same finding appears, which is that AI is a tool and not an answer, and that its value depends entirely on the context it is placed in.

The second wave will be driven by people willing to lead it responsibly, and not by the technology. Technology accelerates and judgment steers, and progress needs both at once.

Sources

Grouped by the section they support.

Opening

Deloitte (2025), The Path to Sustainable Generative AI Value Balances Passion, Pragmatism and Patience, press release on the State of Generative AI in the Enterprise, Q4 report, 21 January 2025, deloitte.com. Survey of 2,773 director- to C-level respondents in 14 countries. Supports the finding that more than two-thirds of respondents expect 30 percent or fewer of their generative AI experiments to be fully scaled within three to six months.

Speed is not direction

Beauchene, Duranton, Kalra and Martin (2025), AI at Work: Momentum Builds, but Gaps Remain, Boston Consulting Group, 23 June 2025, bcg.com. Supports the finding that only 13 percent of employees see AI agents deeply integrated into their daily workflows, and that 51 percent of frontline employees use AI tools regularly.

Companies are measuring the wrong thing

No external source. The three practices of organizations that achieve an effect are the author's synthesis.

Trust comes from understanding

Gillespie, Lockey, Ward, Macdade and Hassed (2025), Trust, attitudes and use of artificial intelligence: A global study 2025, University of Melbourne and KPMG, kpmg.com. Supports the 46 percent of people globally who are willing to trust AI systems, the 56 percent who have made mistakes in their work because of AI, and the 66 percent who rely on AI output without evaluating its accuracy.

The second wave is a mental shift

World Economic Forum in collaboration with Accenture (2025), AI in Action: Beyond Experimentation to Transform Industry, 21 January 2025, weforum.org. Supports the finding that many organizations have run pilots and proofs of concept while scaling them to sustained impact remains a significant challenge. The shift in posture described here is the author's own reading.

Experience is the corrective