Output-Competence Decoupling: How Generative AI Is Creating Illusions of Expertise in the Workplace
Recent research indicates that while AI boosts novice productivity, it offers negligible benefits to experts, creating a workforce capable of generating complex artifacts without the underlying judgment to verify them.
A growing concern within the professional services sector is the emergence of a phenomenon termed "output-competence decoupling." This dynamic occurs when generative artificial intelligence enables staff without formal training to produce work that mimics the quality of experts, yet lacks the underlying expertise or judgment required to verify it. In this environment, the ability to generate a deliverable no longer serves as a reliable signal of the worker's competence, as the output reflects the capabilities of the machine rather than the human operator.
The failure modes of this technology manifest in two distinct ways that are reshaping the workplace. The first involves novices in a specific field producing work that resembles the output of their seniors, often moving faster or appearing more advanced than their actual judgment allows. The second, and arguably riskier form, involves individuals generating complex artifacts in disciplines they have never been trained in. This includes non-engineers building sophisticated data systems or software, resulting in schemas and objectives that are fundamentally flawed from the outset.
Management incentives often prioritise the appearance of momentum and volume over technical accuracy, allowing these flawed systems to persist. A recent case involving Deloitte highlighted the financial stakes, where the firm refunded part of a $440,000 fee after delivering a government report containing AI-hallucinated content. Such incidents underscore the danger of hollowed-out production systems where the institution invests in the illusion of progress rather than substantive value.
Recent studies indicate that leading AI models are approximately 50 per cent more agreeable than human respondents, a trait known as sycophancy. This tendency affirms users even when the affirmation is unwarranted, effectively boosting the confidence of novices while offering negligible benefits to experts. Consequently, workers may become overconfident in their ability to review their own output, unable to detect errors because the tool has already validated their incorrect premises.
The economic landscape of professional work has shifted such that the cost of producing synthetic documents has fallen to nearly zero, while the cost of reading and verifying them has risen. This disparity has led to document inflation, where requirements and status updates are elongated into lengthy, difficult-to-parse artifacts. The slowness that once constituted the real work of learning and refining a craft is now bypassed, leaving a pipeline of future experts that is thinning from both ends.
Experts warn that the competitive advantage for firms lies in work that can be trusted, yet many organisations are quietly converting themselves into content-generation pipelines. As the signal for quality becomes harder to find within a flood of synthetic motion, the reckoning will likely arrive when clients begin to manually review deliverables and discover that the work does not match the invoice.
