Workers spend approximately 6 hours each week directly using artificial intelligence (AI) but dedicate an additional 6,4 hours to preparatory and follow-up tasks. This data was presented at the Vietnam AI Productivity Activation program held on 23/9 in Ho Chi Minh City.

The information was shared by Thanh Nguyen, CEO of Anphabe, citing a report from the Work AI Institute at Glean, in collaboration with researchers from the University of Notre Dame, Stanford, and UC Berkeley. Anphabe, OpenAI's first productivity and capability transformation partner in Vietnam, organized the program.

Thanh Nguyen, CEO and Chief Happiness Officer of Anphabe, spoke at the program on the morning of 23/9. Photo: An Phuong

According to Nguyen, a current paradox is that employees use AI expecting faster work, but "are not sure if they save time."

The act of querying AI represents only a small part of the overall process. Before submitting a request, users must prepare the context, data, and prompt to ensure appropriate input. After receiving results, they still need to review, edit, evaluate, and make decisions. Therefore, only accounting for the AI's response time does not fully reflect actual productivity.

A survey indicates that 28% of AI users spend extra time finding and preparing context, while another 28% spend considerable time adjusting the AI-generated output.

These tasks are known as "boss sitting," where users continuously prepare, guide, monitor, and verify the AI's work.

In Vietnam, an Anphabe survey completed in 8/2026 reveals that 58% of employees use AI for work. However, only 13% of businesses have a formal program for integrating AI into their organization, and 4% feel confident in having established a structured workflow between humans and AI.

The survey also shows that 68% of AI users in Vietnam utilize personal AI tools and pay for them independently. This signifies AI entering businesses through a "backdoor," as employees find tools to support their work before organizations can build governance mechanisms.

Employees might input work-related data, such as product specifications, financial models, sales forecasts, or employee survey results, into personal AI tools. This poses a risk of data leakage. Nguyen noted that most cases do not stem from an intention to disclose information but from the need to complete tasks quickly.

Businesses must change their approach, not just purchase tools

According to Nguyen, the gap between using AI and generating real value from it arises from two main categories of issues.

The first is the AI infrastructure within the enterprise. When using personal AI accounts, the tools cannot access full internal data, leading to "information blindness." AI provides suggestions without sufficient context, forcing employees to manually input data and spend more time. Nguyen believes this problem can be resolved with appropriate technology and data organization.

Next is how businesses organize work with AI. Productivity depends not only on the tools but also on how businesses redesign the collaboration between humans and AI.

Office workers during working hours in Ho Chi Minh City. Photo: Quynh Tran

Mark Jeffrey, a representative from OpenAI, observed that Vietnam has a high rate of AI adoption, more than double the global average. However, most users currently focus on asking questions and receiving answers, whereas AI could be used for data analysis, connecting with internal information sources and tools, and supporting software development.

Jeffrey emphasized that the crucial aspect is not just having AI tools, but ensuring people possess the skills to leverage the technology and translate it into actual work changes. Technology is merely a tool; achieving results requires simultaneous changes in human capability and work organization.

Another paradox is that a single AI-savvy individual may not make the entire organization more proficient in AI. An employee might discover effective ways to use AI independently, but if this approach remains confined to a personal account, the business struggles to scale it.

Nguyen stated that businesses cannot simply provide AI tools and expect employees to become more efficient automatically. The workflow surrounding AI must also evolve.

Based on the AI productivity capability framework developed by Anphabe and OpenAI experts, users are categorized into four levels. At the first level, they primarily engage in question-and-answer interactions and assign simple tasks like translation, summarization, and information retrieval.

At higher levels, AI connects with organizational data and real-world problems to analyze, challenge assumptions, identify blind spots, and propose solutions, with humans making final decisions.

The next level involves identifying repetitive tasks for automation using AI agents. Even higher is standardizing effective methods into processes and documentation for training and scaling across the organization.

Anphabe's data indicates that 82% of AI users in Vietnam remain at the initial level. The Anphabe CEO believes the bottleneck in transitioning from basic AI use to generating results largely lies in data and context. For AI to produce optimal output, it needs quality data and connection to the business's real-time, evolving information.

Regarding the "backdoor AI" situation, Nguyen argued that businesses cannot prevent AI from entering the organization simply by banning external tools. Instead, they need to create a secure AI environment with appropriate data and context for employees to use.

Furthermore, businesses should identify effective individual approaches, standardize them into processes, and share and train others to adopt them. The goal is not to transform all employees into "AI power users," but to elevate effective individual practices into a collective organizational capability.

From a management perspective, Nguyen asserted that AI only generates productivity when all three elements—technology, work, and people—are simultaneously activated; if one element is zero, the added value is also zero. Therefore, businesses need to shift their focus from "which AI tools to invest in, for how many people" to "how will AI change work processes to create real productivity."

Le Tuyet