Artificial intelligence is often compared to human intelligence — but that comparison might be misleading. A growing number of experts now describe AI as having what’s called “jagged intelligence,” meaning its abilities are uneven and unpredictable.

In some areas, AI performs at an incredibly high level. For example, systems developed by OpenAI and Google recently solved complex problems from the International Math Olympiad, showcasing near-expert mathematical skills.

Yet in everyday reasoning, these same systems can fall short. A simple real-world question — like whether to walk or drive a very short distance — has confused AI tools, highlighting their lack of common sense.

The term “jagged intelligence” was popularized by Andrej Karpathy, who explained that AI can excel in one task while failing completely in another — often in ways that are hard to predict. Unlike humans, whose abilities tend to develop in a more balanced way, AI improves in sharp, uneven jumps.

This idea is becoming increasingly important when discussing the future of jobs. Rather than replacing entire professions, AI is more likely to automate specific tasks within those jobs. For instance, coding tools like those from Anthropic or OpenAI can write code quickly, but they still struggle to understand the broader structure of complex software systems.

Economists suggest this means many workers won’t be replaced outright — instead, their roles will evolve. Tasks that AI can handle will be automated, allowing humans to focus on more strategic and creative work.

Another limitation comes from how AI learns. These systems rely heavily on patterns found in online data, which represents only a fraction of real-world human knowledge. As a result, AI is strong in digital tasks like writing, coding, and answering questions, but weaker in areas that require physical understanding, long-term planning, or true creativity.

Despite these gaps, AI is improving rapidly. New training methods, such as reinforcement learning, are helping systems get better at structured tasks like mathematics and programming. However, fields where success is harder to define — such as philosophy or creative writing — remain challenging.

Ultimately, the concept of jagged intelligence offers a more realistic way to understand AI. It shows that the technology is neither fully human-like nor universally capable — but something entirely different, with strengths and weaknesses that will shape how we work in the future.