Survey: 5 stages to AI coworker — why just 'chatting' is obsolete
tencent and universities published a survey mapping five stages of ai evolution from chatbots to digital colleagues. the paper argues persistent work environments and reusable 'skills' are required to transform models from answering questions to reliably finishing tasks.
researchers from tencent's youtu lab and various universities released a survey paper outlining five stages of ai development, moving from basic chatbots to autonomous digital colleagues. the paper identifies a shift from fast 'system 1' text generation to slow 'system 2' reasoning, citing models like openai's o1 and deepseek-r1 that use extended compute for verification and self-correction. it argues that simple tool calling is insufficient for reliable agents, introducing the 'openclaw era' concept where models operate in persistent environments containing files, sessions, and logs. the core proposal combines this persistent workspace with 'skills'—reusable bundles of operational knowledge similar to anthropic's agent skills—to bridge the gap between intent and finished work. this framework suggests future training will move beyond instruction-response pairs to focus on verifiable task completion in complex, stateful environments.