Generative tools such as ChatGPT, Gemini, and Claude have made artificial intelligence (AI) highly visible in everyday work. Millions of people now use such tools to draft emails, summarize notes, translate text, or clean up presentations. This creates a powerful impression that AI has already entered workplaces at scale and that major labor-market effects are imminent. Yet the visibility of AI use and its transformative impact on work are not necessarily the same thing.
The forms of AI adoption most likely to transform productivity and employment are not necessarily the most visible ones: workers occasionally using a chatbot in a browser to save a few minutes on a document. Rather, they involve enterprises – across services, manufacturing, and agriculture – redesigning workflows, integrating AI into operations, and optimizing how work is allocated across teams. It is these organizational changes, rather than the occasional use of AI tools by individual workers, that are most likely to drive substantial productivity gains and labor-market effects. That kind of adoption is difficult and takes time; it requires organizational capability and high-quality data infrastructure, not just access to the technology or a subscription to a service.
This matters especially in the Asia-Pacific, a region whose labor markets are heavily shaped by informality and small-scale economic activity. Nearly 66 percent of employment in the region is , reaching up to 85 percent in sub-regions such as South Asia. Self-employment, micro-enterprises, and small enterprises together are estimated to account for 54 percent of total employment in East Asia, South-East Asia, and the Pacific, and 91 percent in South Asia. Such enterprises often face greater barriers to adopting transformative AI tools, including limited resources to invest in digital infrastructure, workforce skills, and the organizational changes needed to integrate AI into business processes.
While AI may lower the entry barriers for some – for example, allowing informal vendors to use simple tools for basic accounting – meaningful productivity gains require more than just a smartphone. They require a level of organizational capability that many informal and small enterprises currently lack. Effective adoption often demands ongoing investment in skills, data systems and business-process redesign as AI technologies evolve. These requirements are easier for larger enterprises to absorb, while leaving smaller firms, operating with limited resources and limited managerial capacity, struggling to keep pace.
That may be one of the key labor-market challenges related to AI in significant parts of the Asia-Pacific region.
Emerging evidence suggests that these barriers are already visible in practice. New survey data from Singapore, one of the region’s most digitally advanced economies, show that 71.5 percent of enterprises have not yet adopted AI; the percentage is 76 percent for firms with fewer than 25 employees, compared with only 24 percent among firms with more than 500 employees. A survey among employer and business membership organizations (EBMOs) in the Asia-Pacific suggests that on average nearly three quarters of EBMO members are reportedly still exploring AI or at an early stage of adoption.
If this trajectory continues, the main near-term effect of AI may not be drastic job loss, nor a broad-based productivity surge. It may instead be a widening gap between a relatively small set of frontier enterprises that can absorb AI into their operating model and a much larger set of enterprises that cannot. In that sense, AI may first deepen existing inequalities before it transforms employment at a larger scale.
This perspective helps explain a puzzle already visible in labor market data. Even in occupations thought to be highly exposed to AI, early evidence from labor force surveys in the Asia-Pacific region does not yet suggest that employment trends have shifted dramatically. Similarly, AI-driven gains are not yet reflected in broader productivity trends across the region. This highlights that potential exposure tells us only where AI could theoretically matter, but not whether enterprises are actually adopting it in ways that change production processes, productivity or employment.
That distinction matters especially in the Asia-Pacific, because the region’s development challenge has never been only about access to frontier technology, but about diffusion – ensuring that a much wider range of firms have the capabilities, infrastructure, and organizational conditions to use new technologies effectively. New technologies often arrive first in globally connected, large formal enterprises, while the broader economy lags behind. The result is likely to be uneven productivity growth, uneven job quality, and wider divergence between leading enterprises and everyone else.
AI appears to be following the same pattern.
If this trajectory continues, then much of today’s debate may be incomplete. Debates should not focus solely on which tasks AI may technically be able to automate. An equally important question is whether AI can diffuse into the segments of the economy where most of the region’s 2 billion workers are actually employed.
Will AI also reach smaller and informal enterprises? Will all entrepreneurs and workers have the skills to integrate AI in their business models and to use it meaningfully and systematically? Will the digital infrastructure allow all enterprises and workers to benefit from the adoption of AI systems?
To ensure that the answer to these questions is “yes,” policies cannot focus only on frontier innovation or the occupations most exposed to automation. They must also address the conditions that allow AI to diffuse more broadly across the economy, including digital infrastructure, skills development, and support for technology adoption among smaller enterprises. Alongside broader policy measures, social dialogue between employers and workers at the workplace level can help foster transparency in the adoption of AI systems and support alignment with both business objectives and workers’ capabilities and expectations. In addition to developing more powerful systems, the challenge will be to ensure that a much wider range of firms and workers can use them productively.
Otherwise, the Asia-Pacific may end up showing that the key divide in the AI era is between enterprises that can reorganize production to integrate AI in ways that augment work and raise productivity, and those that cannot – leaving many workers in low-productivity settings largely untouched by its benefits. In other words, AI may raise the ceiling for some enterprises without lifting the floor for most workers.
