A survey paper argues that AI systems won't become reliable coworkers until they finish entire tasks in persistent work environments instead of just generating answers. The key lies in reusable 'skills' that allow AI to perform complex, multi-step actions. This marks a significant evolution in how AI is perceived and utilized in professional settings.

A research team from Tencent's Youtu Lab and several Chinese universities maps the shift 'from chatbot to digital colleague' along two dimensions in a new survey paper: the cognitive core and tool-assisted task execution. The central question is no longer how a model produces a better answer, but how it reliably turns intent into finished work, according to the researchers.

In the chatbot era, models mostly generated text fast. They stored language patterns and facts in their parameters, then wrote answers in one pass, token by token, following the most likely continuation without checking intermediate steps or searching for solutions. This approach, while efficient, lacks the depth required for complex tasks.

The transition to a thinking-LLM era, initiated by models like OpenAI's o1 and Deepseek-R1, emphasizes slow thinking and reasoning. These models are designed to handle complex tasks by breaking them into smaller steps, evaluating outcomes, and adjusting strategies as needed. This shift is expected to redefine how AI integrates into workflows and collaborates with humans.

The implications of this shift are significant for organizations relying on AI for productivity. As AI systems evolve to perform tasks rather than just provide answers, they will require more integration with existing tools and workflows. This may lead to increased costs, potential vendor lock-in, and the need for new governance frameworks to ensure AI is used responsibly and effectively.