AI Use: Complex Tasks Are Delegated, Routine Remains

The reasoning of many bosses sounds plausible: AI handles routine tasks, employees focus on complex problems. But reality looks different, as a new investigation by AI developer Anthropic shows (PDF). In practice, it is primarily the complex tasks that are delegated to artificial intelligence, for which it has a significantly lower success rate of 66 percent compared to simple tasks. The consequence: Many jobs are “deskilled,” as the researchers call it. The complex parts disappear; the routine work remains.

The study is based on two million anonymized usage data from November 2025, which were analyzed by AI. For the first time, Anthropic systematically measured not only what Claude is used for but also how successfully.

“Deskilling” of Professions

The study cites technical writers and travel agents as examples of this “deskilling.” With the help of AI, the technical writer analyzes new developments to determine the need for revisions or has already published materials reviewed to identify revision needs. While they delegate these complex tasks, they prefer to handle sketching or production themselves. In travel agencies, tasks such as requesting and printing tickets or processing payments are preferably left to humans. However, AI assistance is sought for difficult tasks such as planning travel routes or cost calculations.

This is surprising at first glance because errors in the results of complex tasks are often harder to detect than in simple tasks. And according to Anthropic's figures, the success rate for simple tasks is 70 percent, while it drops to 66 percent for complex cases. The fact that users accept this is likely due to the time savings. Tasks that would take people about three hours without AI can be completed with AI in about 15 minutes, Anthropic claims.

AI Widens the Gap Between Rich and Poor

Another finding from the study is that AI could further increase inequality between richer and poorer countries and their respective living conditions. In rich countries with higher per capita income, AI is used versatilely and also for private purposes, for example in the USA, Japan, and South Korea. In poorer countries, its use is concentrated on learning and specific work tasks.

Such thoughtfulness about the side effects of one's own products seems unusual at first glance in the AI industry. However, Anthropic has already attracted attention several times in the past by counteracting the euphoric tone that usually characterizes Big Tech with a degree of self-reflection. And often with the success that the smaller competitor to OpenAI (ChatGPT) and Google (Gemini) made headlines with it.

(mki)