Over a year ago, the 40-year-old woman’s company sacked half its employees after integrating artificial intelligence into its operations. The team shrank but the workload remained very high, and so the employees resorted to use of AI.
AI-generated stuff often looks good on the surface but carries huge risks.
In late 2025, an employee used AI to write an advertisement, and ended up conflating technical specifications of two different car models. The error forced the company to pay compensation to the client and lose 30% of the contract value.
After this, Dung banned her staff from using AI at work, but the resultant slow progress forced her to lift the ban in just a week. Now, to ensure quality, she spends an extra five to 10 minutes on each report, verifying every point and cross-checking data.
In the past, she used to get only 10 of them in a day, and had plenty of time left to meet clients. Now, she has to go through 40 daily, which needs hours and hours and forces her to take work home every night.
“The company cut labor costs thanks to AI, but I’m the one picking up the extra load,” she says.
|
Office workers work in an office in HCMC. Photo by Quynh Tran |
This phenomenon of managers cleaning up after their staff is also occurring at Minh Tu’s import-export company in Hai Phong City. Since adopting AI, the company no longer needs to hire translators, and the time required to generate market research reports has dropped from months to a few days.
But there are issues.
Tu once had to translate a partner’s 10-page document herself since the summary an employee had generated using Gemini omitted core details and made up parameters. On another occasion, an employee submitted a market research report containing fabricated figures generated by ChatGPT simply because it “sounded coherent.”
Instead of focusing on strategic planning, Tu now has to read original foreign-language documents, cross-checking every figure and citation.
“The verification step seems insignificant, but takes up a massive amount of time,” she says.
Dung and Tu’s stories reflect a growing phenomenon known as “workslop”, referring to AI-generated products that look polished but lack substance or contain false information.
The term gained attention following a survey of over 1,000 U.S. workers published in Harvard Business Review in late 2025. The report revealed that 40% of employees had received “AI slop” within the past month, mostly from colleagues (40%), subordinates (18%), and even superiors (16%). Dealing with this low-quality content took up 15% of their working time, causing 42% of workers to lose trust in their colleagues and 53% to feel frustrated.
In Vietnam, data published at the Vietnam AI Productivity Activation event in HCMC on Sept. 23 showed that workers using AI save about 11 to 13 hours per week, but spend an additional 6.4 hours preparing data, editing, and verifying results.
In some cases, this creates a “productivity paradox,” where the time spent supervising and correcting AI is greater than the time saved by using it. This added workload is often referred to as “AI babysitting”, a state where humans must continuously guide, supervise, and correct AI outputs.
Notably, a survey by Glean’s Work AI Institute in collaboration with top U.S. universities such as Emory, Stanford, UC Berkeley, and UCL found that 69% of employees admitted to submitting AI-generated work without reviewing or checking it first.
Pham Lam Tung, managing director of an AI training center, says many companies have paid a heavy price because their employees let AI do the whole work.
At a bank in HCMC, he witnessed a branch director having to redo a financial report after discovering a subordinate fed a blurry scan into AI, causing it to misread the data and make up figures to fill in the blanks.
In Hanoi, another bank’s PR team printed large-scale posters of an AI-generated image. It was only after the posters were put up that they noticed the figure in the image had misaligned eyes.
“It’s not just employee—many bosses also misuse AI,” Tung says.
“A startup founder in the northern Bac Ninh Province regularly shared ideas with ChatGPT and received glowing praise. As a result, he shifted strategies every week, forcing his team to constantly revise plans.”
According to Tung, “AI slop” stems from two causes: employees’ lack of AI-prompting skills while using free versions with limited features or do only the bare minimum due to laziness or burnout.
 |
|
University students learn about AI at an event in HCMC. Photo by Bao Lam |
Hong Anh, 28, an advertiser in Hanoi, is part of the second group. Her workload, which doubled after a round of layoffs in 2025, forces her to generate scripts with AI in order to hit her KPI targets.
“Between getting fined for missing deadlines and submitting a smooth-sounding draft, hoping the boss won’t pay too much attention, I choose to do the latter,” she admits.
This approach incurs hidden costs: Companies spend resources fixing errors and face legal risks and data leaks; while managers lose trust and employees gradually lose their critical thinking skills.
Tung says that some banks and companies have decided to completely ban the use of AI to prevent risks. However, he considers this ineffective because employees can access AI with their mobile network or use it when working from home.
Instead of bans, companies need to combine AI training with clear ethical and data-security standards, he says. “Don’t start with the tool, start with accountability.”
It is echoed by Dat Le, CEO of a software and AI firm in HCMC, who says that as a management principle, when a manager receives a low-quality report, they should send it back for the employee to review themselves instead of doing the work for them.
“The manager should only care about the final output, regardless of whether the employee used AI or not,” he adds.
But then, companies cannot expect high-quality output without investment. On a monthly salary of VND10 million (US$386), it is unrealistic to expect employees to spend VND500,000 to 1 million out of their own pockets for AI subscriptions. Their use of free AI tools without guidance, combined with a hands-off mindset, inevitably leads to flawed results.
In working sessions with business owners, Dat frequently emphasizes that as “AI personnel” gradually replace humans in hard skills like coding, design, or content creation, the core value of human workers will lie in qualities machines cannot replicate such as emotion, meticulousness, empathy, and judgment.
“To work with AI effectively, users need to think of themselves as its manager and constantly verify the output. The tool isn’t at fault for a bad product, the fault lies in the negligence of the person using it,” he states.