Imagine you’re a technical writer facing a deadline. You could spend hours analyzing new software developments to determine what needs updating, or you could ask an AI to do it in minutes. According to a groundbreaking study from AI developer Anthropic, most professionals are choosing the latter – and it’s reshaping the workplace in unexpected ways.
The Deskilling Paradox
Anthropic’s analysis of two million anonymized usage records from November 2025 reveals a counterintuitive trend: workers are delegating complex, high-value tasks to AI while keeping routine work for themselves. The study shows AI achieves a 66% success rate on complex tasks compared to 70% on simple ones, yet users overwhelmingly choose to offload the difficult work. This creates what researchers call “deskilling” – where challenging aspects of jobs disappear while mundane tasks remain.
“The calculation many managers make seems logical: AI handles routine work, employees focus on complex problems,” the study notes. “But reality looks different.” Technical writers, for instance, use AI to analyze new developments and determine revision needs, while continuing to handle sketching and production themselves. Travel agents let AI plan routes and calculate costs but keep ticket printing and payment processing.
The Productivity Trade-Off
Why would workers accept lower accuracy on important tasks? The answer lies in dramatic time savings. Anthropic claims tasks that would take humans three hours can be completed with AI in about 15 minutes. This efficiency gain comes with hidden costs, however. As AI operations manager Bernard Meyer warns, “AI should give people more control over how they spend cognitive energy instead of removing it. But that requires discipline.”
The deskilling phenomenon raises questions about workplace creativity. Research cited by ZDNET shows that moderate boredom and repetitive tasks can actually foster creative insights. A study in Scientific Reports found brief episodes of boredom may trigger cognitive reorganization, enabling deeper engagement with material. Lacey Kaelani-Dahan, founder of software company Metaintro, observes: “A large number of our best product ideas have come from engineers doing the same repetitive data validation work over and over again, where they notice patterns that would lead to larger insight.”
Security and Privacy Concerns
As AI agents gain more access to workplace systems, security vulnerabilities become increasingly concerning. A recent report revealed a security flaw in Anthropic’s Claude Cowork that allows hackers to exfiltrate files from users’ local folders through indirect prompt injection attacks. Security firm Promptarmor identified the vulnerability, which exploits isolation flaws in Claude’s code execution environment.
British software developer Simon Willison, who coined the term “Prompt Injection,” criticized inadequate warnings: “I don’t think it’s fair to tell ordinary non-programmers to watch for ‘suspicious actions that might indicate a prompt injection’!” Meanwhile, privacy-focused alternatives are emerging. Signal founder Moxie Marlinspike recently launched Confer, a chatbot that uses confidential computing and trusted execution environments to protect conversations from being read by providers or authorities.
The Global Inequality Factor
Anthropic’s study reveals another troubling trend: AI may be widening global inequality. In wealthier countries like the U.S., Japan, and South Korea, AI is used diversely for both work and personal tasks. In poorer nations, usage concentrates on specific work tasks and learning. This digital divide comes as Anthropic expands aggressively into emerging markets, appointing former Microsoft India managing director Irina Ghose to lead its Bengaluru expansion.
Ghose notes growing demand for “high-trust, enterprise-grade AI” in India, where Claude app downloads increased 48% year-over-year in September. “AI tailored to local languages could be a ‘force multiplier’ across sectors including education and healthcare,” she says. However, this expansion occurs alongside competitive pressure from OpenAI, which introduced ChatGPT Go, an under-$5 plan for Indian users.
Finding the Right Balance
The key challenge for businesses lies in balancing AI’s efficiency gains with human cognitive needs. Shawn Spooner, global chief technology officer at advertising agency Billups, suggests: “A better paradigm would be to aim to keep our folks in flow state for a few multi-hour blocks each day, where they are the most productive and the most happy, stitched together with more mundane tasks to give their day a more natural ebb and flow.”
Debra Andrews, president of consulting firm Marketri, warns against over-optimization: “There’s a risk in assuming that every recovered minute should be filled with more high-intensity work. If teams are constantly pushed from one cognitively demanding task to the next, creativity and strategic clarity actually suffer.”
As AI continues to evolve, the question isn’t whether to use it, but how to integrate it thoughtfully. The deskilling trend identified by Anthropic serves as a cautionary tale: without careful planning, we risk creating workplaces where humans handle the boring parts while machines tackle the interesting challenges – exactly the opposite of what most innovation promises.