Despite rapid and widespread adoption of AI tools, a notable phenomenon has emerged: the majority of companies report not yet seeing significant earnings gains from their AI investments. McKinsey has termed this the Gen AI Paradox.
This discrepancy is largely attributed to the fact that most companies are still operating in copilot mode, where AI primarily assists with prompt-based tasks, rather than fully leveraging more advanced agentic AI systems capable of managing entire processes autonomously.
The value derived from copilots is often limited to specific tasks such as writing faster, organising data, or summarising meetings. Whilst these benefits are real, they are frequently difficult to quantify and rarely scale across an entire organisation, making it challenging to demonstrate significant, measurable business impact.
In contrast, agentic AI systems demonstrate the potential to condense a month's worth of work into a single day.
Lenovo's engineering teams, for instance, experienced up to 15% improvement in code quality and speed after implementing AI agents, and their customer support AI agents resolved the majority of inbound queries, reducing response times by up to 90%.
Recent survey data from enterprise research firms reinforces the challenge.
Only 8.6% of companies report having AI agents deployed in production, whilst 14% are developing agents in pilot form and nearly 64% report no formalised AI initiative at all. However, the share of organisations with deployed agents nearly doubled in just four months in late 2025, suggesting that the enterprises investing through 2026 are increasingly those with the operational discipline to move past experimentation and into repeatable, scaled use cases.
Initial productivity dips and long-term gains
The adoption of AI frequently leads to an initial, measurable but temporary decline in productivity, a phenomenon often described as a J-curve effect.
This initial dip is typically followed by stronger long-term growth in output, revenue, and employment.
Successfully integrating AI often necessitates additional investments in data infrastructure, comprehensive staff training, and fundamental redesign of workflows. Older, more established firms tend to experience more pronounced short-term losses due to their entrenched routines, layered hierarchies, and complex legacy systems.
This productivity paradox highlights that AI implementation is a transformation cost, not solely a technology expense.
Organisations must budget for and anticipate these short-term productivity losses as an unavoidable part of the journey towards AI-driven efficiency and growth. The critical point for investors and decision-makers is that this J-curve is well-documented and predictable. Companies that plan for it, with adequate budget, realistic timelines, and proper change management, consistently emerge stronger on the other side.
Challenges to AI adoption and trust
The path to widespread enterprise AI adoption is fraught with significant challenges, including data quality, integration complexities, workforce resistance, and ethical concerns.
The effectiveness of AI models is inherently tied to the quality of the data they are trained on. Poor data quality can lead to unreliable insights and flawed decision-making, and 42% of respondents in major surveys express concern about insufficient proprietary data to customise models.
Algorithmic bias presents a critical ethical concern. This bias often stems from non-representative or historically inequitable training datasets, which can lead to unequal treatment and erosion of trust.
Integrating AI-driven workflow agents seamlessly with existing tools, APIs, and legacy systems remains a significant challenge, and many traditional systems lack modern, AI-agent-friendly APIs. Data maturity is another major limitation: siloed data, missing metadata, or outdated records can lead to AI agents producing inaccurate outputs requiring human intervention.
AI adoption also frequently encounters significant workforce resistance, stemming from unfamiliarity with new technologies, perceived skill gaps, fear of job displacement, and disruption to established routines.
Gartner's observations highlight that deployment does not equal adoption, emphasising that enterprises frequently underestimate the significant change-management effort required for successful AI implementation.
