After writing a blog post about work-life balance for her employer, Emilie Schario did what many people now do before publishing anything online: She pasted her draft into an AI tool in hopes of getting back a stronger post.
Yet rather than take the revision it produced at face value, Schario said she spent close to half as much time reviewing the new version as she did writing the original — and for good reason. The AI added a sentence saying that she had recently blocked time on her calendar to attend her daughter’s school play.
“I don’t have a daughter, and there was no school play,” said Schario, chief operating officer at Kilo Code, a remote AI coding startup, and the mother of three young boys.
AI tools are helping workers complete all sorts of tasks faster than ever before, yet many are discovering a drawback. The output still requires careful review for errors and hallucinations, cutting into the time the technology is meant to save.
New survey data provides a sense of just how much. Nearly 40% of AI’s value is lost to rework and misalignment, and only 14% of employees consistently get clear, positive outcomes from the technology, the survey found.
The survey was conducted in November by Hanover Research on behalf of HR and finance software provider Workday. Respondents included 1,600 leaders and 1,600 full-time employees from companies worldwide with annual revenues of $100 million or more.
AI still nets out to be a time-saver, and editing the results will likely become less of a chore in the future as the technology advances — and as workers receive more training on how to write prompts and apply critical thinking to AI-generated work, said Workday executive Aashna Kircher.
“We’re seeing a need for organizations to better enable their people to evaluate the output and make the right decisions in terms of how it’s used,” she said.
How training could help
In Workday’s survey, 66% of leaders cite skills training as a top priority, yet only 37% of employees facing the most AI rework say they’re getting it. The findings also show that fewer than half of employee job descriptions have been updated to reflect AI capabilities, leaving workers to balance faster AI-driven output with the same expectations around accuracy, judgment, and risk.
Other studies also suggest that AI outputs routinely require human intervention and cannot be trusted outright. A global survey of 2,000 CEOs found that only a quarter of AI efforts had delivered the returns the leaders had expected. It was conducted between February and April of last year by the IBM Institute for Business Value and Oxford Economics.
Similarly, 95% of organizations reported no measurable ROI from AI, according to an MIT study based on reviews of publicly disclosed AI initiatives and executive interviews between January and June of last year.
Schario, who works for Kilo Code from her home in Columbus, Georgia, isn’t giving up on AI. As someone who finds writing as unpleasant as folding laundry, she said the speed at which these tools work outweighs the need to have to carefully check for errors.
“I think where people get themselves in trouble is that they take that output of the AI agent, they don’t review it closely, and they just kind of pass it on,” she said. “At the end of the day, you are still responsible for your output, whether it was generated by an AI agent or not.”
Read the original article on Business Insider
The post AI is speeding up work, but fixing its mistakes eats away at productivity gains appeared first on Business Insider.
