How to destroy productivity with AI
Artificial intelligence firms promised a historic, seismic boom in productivity for businesses and economies writ large. The hundreds of billions of dollars being poured into the technology and its infrastructure speak to how strong the belief in that promise is.
So far, however, when it comes to businesses, the gains are patchy – and in many cases are not gains at all, but in fact losses.
One of the culprits responsible for this is encapsulated in a term now familiar to many of the companies that have tried to implement AI into their operations: AI workslop.
If you haven’t experienced it, you have most likely heard of it.
The Harvard Business Review defines AI workslop as “AI-generated work content that masquerades as good work, but lacks the substance to meaningfully advance a given task.”
Generative AI tools enable workers to produce polished-looking reports, slide decks, summaries of long research papers, and usable code. But in many cases, the AI-created products lack critical information, are incoherent or contain completely inaccurate content altogether.
Insufficient training data or biases in the data used to train large language models can often lead to hallucination, whereby the AI generates false assumptions and incorrect or misleading results – creating potentially huge liabilities for companies.
This leaves the recipient of the product having to comb through it to correct errors, or redo the work entirely.
A survey conducted by digital coaching BetterUp in partnership with the Stanford Social Media Lab in September found that 40% of U.S.-based desk workers reported receiving workslop in the prior month, and that it took an average of two hours to resolve each incident where AI-generated work produced errors.
Translated into dollar terms, the research found that these mistakes cost $189 per employee per month, amounting to a whopping $9 million in annual costs for a 10,000-person company.
Now take the fact that 95% of knowledge work has yet to be automated. This presents an almost terrifying scope for the degradation of work quality across the entire economy as AI gradually penetrates more and more sectors.
How do we prevent this? Human-in-the-loop accountability.
At AppliedAI, human oversight and transparency will always have a place in the future of work. With the Opus workflow automation platform, every task is tracked, auditable, and requires either agentic or human review. Opus’ audit log shows who checked and signed off on each task.
As LLMs develop, meanwhile, businesses shouldn’t give up on implementing the technology for work – but rather approach it with more discernment, says AppliedAI CEO Arya Bolurfrushan:
“You can’t throw the baby out with the bathwater. Supervised automation means automating many of the tasks, but the human should be the final check. And every piece of writing should have a human name next to it. So someone has looked at it. And if there’s a mistake, that person is liable.”
“Let’s say a piece of work is 100% – 90% of it should be done by prompting, but the last 10% is the cleanup that has to be done.”
AI is a partner to human intelligence, not a replacement for it. And in order to uphold high quality standards, there must be humans involved in the process that remain accountable – a machine will not suffer consequences of poor work or taking shortcuts.
For now, the “GenAI divide” – defined by MIT as high adoption of generative AI tools but low transformation – highlights the challenge ahead for true integration and transformation of industries by the technology.
A July 2025 report from the MIT Media Lab’s NANDA initiative found that despite as much as $40 billion in enterprise spending on generative AI, 95% of organizations are seeing no measurable business return.
“Just 5% of integrated AI pilots are extracting millions in value, while the vast majority remain stuck with no measurable P&L impact,” the report read.
Development and engineering across countless startups and major firms are powering ahead to change this, aided by bullish capital flows from investors.
But understanding business customers’ day-to-day operations, improving contextual learning, ensuring seamlessness of workflows, delivering productivity at scale, and fine-tuning the relationship between AI advancements and irreplaceable human judgement will ultimately be what sets apart the winners from all the rest.