From Acting to Accomplishing: A Survey of Language-Agent Autonomy in Digital Work

PhD Qualifying Examination


Title: "From Acting to Accomplishing: A Survey of Language-Agent Autonomy in 
Digital Work"

by

Mr. Baixuan XU


Abstract:

Large Language Models (LLMs) are catalyzing a paradigm shift in digital work, 
evolving from task-specific automation tools into increasingly autonomous 
agents and fundamentally redefining the division of labor between human 
operators and artificial intelligence. This survey systematically charts this 
progression, placing a central focus on the changing roles and escalating 
autonomy of language agents. Through the lens of the work 
cycle-specification, planning, execution, tracking, verification, and 
acceptance-we introduce a foundational three-level taxonomy: Acting, 
Enduring, and Accomplishing, to delineate which stations of the cycle the 
agent owns at each level. We further identify pivotal challenges and future 
research trajectories, including calibrated completion judgment, 
specification fidelity at scale, autonomy-aware evaluation design, training 
toward autonomous task completion, and oversight of task-owning agents. 
Overall, our examination reveals that five of the six stations are 
progressively transferring to the agent, while the sixth-acceptance, the 
ability to stand behind one's own completion claim with evidence sufficient 
for third-party review-remains unconquered and defines the central frontier 
for future research


Date:                   Wednesday, 29 July 2026

Time:                   9:00am - 11:00am

Venue:                  Room 5506
                        Lift 25/26

Committee Members:      Dr. Yangqiu Song (Supervisor)
                        Prof. Raymond Wong (Chairperson)
                        Dr. May Fung