Is AI About to Automate Every Office Job? | AI Reality Check

Mustafa Suleyman’s prediction that AI will fully automate most knowledge work within 18 months lacks empirical support and contradicts the consensus among other industry leaders. While models like GPT-4 demonstrate impressive capabilities, they function as "story completers" rather than autonomous agents, hitting significant scaling walls that prevent rapid, generalized automation. True progress in fields like software engineering stems from specialized "coding harnesses"—human-engineered systems that integrate AI into specific workflows—rather than inherent leaps in model intelligence. Because most knowledge work lacks the structured data and verifiable success metrics found in programming, these tools remain limited to narrow, administrative tasks like summarization and data formatting. Ultimately, the hype surrounding total economic disruption serves corporate interests more than it reflects the current, incremental reality of AI development, which requires focused, human-led integration rather than wholesale replacement.