According to God of Prompt on X, citing Berkeley Haas researchers Aruna Ranganathan and Xingqi Maggie Ye, an eight-month embedded study in a 200-person tech company found that companywide access to AI tools increased pace, widened role scope, and extended work hours, resulting in higher, not lower, workloads (as reported by Berkeley Haas via the X thread). According to the study summary shared by God of Prompt, patterns included task expansion across roles, blurred time boundaries due to near-zero task start friction, and cognitive overload from parallel AI agent use. According to God of Prompt, a 2024 Upwork study reported 77% of AI users said AI increased their workload, and nearly half were unsure how to meet expected productivity gains, reinforcing the Berkeley findings. As reported in the X thread, the researchers call the reinforcing loop workload creep—AI speeds tasks, expectations rise, reliance on AI grows, scope expands, and workload intensifies—creating short-term momentum but long-term strain and burnout risk. According to the Berkeley Haas recommendations summarized in the X post, teams should adopt AI Practice: structured reflection intervals, explicit do-not-expand task lists, and predefining scope and done criteria to capture AI gains without unsustainable escalation.