'AI trash' in the workplace
Nguyen Hai, project manager of a software company providing AI transformation platforms for businesses in Hanoi, said that before the trend of vibe coding, or programming by giving commands to AI, appeared, engineers had to type each line of code. Although it takes more time, this method helps control from idea to implementation.
"Now we often encounter situations where junior personnel (new to the profession) use AI but don't check it and don't understand what's in their code," he said. The product may temporarily run, but the programmer does not understand the system design and is unable to describe or edit it.
According to Mr. To Manh Hoang, CEO of Datapot - a unit specializing in training and providing data and AI conversion solutions for businesses, poor quality but neatly packaged AI products are appearing in many businesses and industries.
He cited a survey by human resources company BetterUp and Stanford University, starting in September 2025 and still continuing to collect, showing that 40% of employees have received "AI junk" - products or reports that look complete but have no value to the overall work.
On the other hand, half of survey participants admitted to sending unvetted AI-generated content to colleagues. These contents mainly appear in the field of professional services such as law, finance, human resources, marketing or technology.
From experience in training and transition support for more than 150 businesses in Vietnam, Datapot representatives recommend two ways to control "AI garbage": do not depend and always verify.
Mr. Hoang assessed that excessive dependence is a problem many junior-level personnel encounter. This group easily falls into the situation of not fully understanding the business, giving orders to AI without enough expertise to verify the results.
"Dependency leads to a decline in skills, 'transfer' of reasoning and perception to AI models," he said. Therefore, personnel in the new career stage need to learn independently, separately from AI to develop their capabilities before using support tools.
Experts also recommend that all AI output should be "default flawed" and need to be adjusted. According to the annual Developer Survey published by the Stack Overflow community in July 2025, conducted on more than 49,000 programmers, 66% rated the most frustrating thing when using AI as "approximate results, but not exact". If not reviewed, those errors can accumulate into technical debt (a problem that arises because the programming solution is not optimal and will have to be repaired or upgraded later), affecting the quality of products and services.
From a management perspective, businesses need to have regulations on AI, even if there are no large-scale transformation plans. "At a minimum, it is necessary to provide guidance to personnel on the scope of work allowed to use AI and how to use it," Mr. Hoang said.
He further noted that the number of businesses with AI processes and regulations is not commensurate with the rate of 66% of working people in Vietnam using AI at work, according to data from the Annual Report on Artificial Intelligence in Vietnam published by Hanoi National University in December 2025.
Sharing the same opinion, Ms. Pham Dieu Linh, Human Resources Director of Synnex FPT, said that organizations need to expand the scope of management not only of people but also of the AI they use at work.
First, it is necessary to identify tasks and tasks in the organization that can be applied to AI. From there, build processes and stages that personnel can coordinate with AI or must directly implement, relevant policies and evaluation indicators.
"When businesses want to use AI to improve performance, they need to determine which process to start with, the responsibilities of people in the process and the level of improvement in output results," Ms. Linh emphasized. That way, the use of AI can be controlled and sustainably replicated within the organization instead of being limited to individual, spontaneous use by individuals.
The report of recruitment platform TopCV in December 2025 also shows that more than 59% of businesses said that in 2026 they will invest in AI skills guidance for personnel. Meanwhile, only 38% will build AI-integrated workflows and 7% plan to build security and compliance processes.
"Most businesses are investing more in training employees to use it at the individual level than building a system," Ms. Linh acknowledged.
Meanwhile, Mr. Nguyen Hai, project manager in Hanoi, said the company had to quickly develop new regulations, allowing engineers to use AI but had to reevaluate the output results for each command line.
"If technical debt is large, AI cannot perform new tasks and engineers cannot understand the system to intervene," he said.
Nam Nguyen