LobsterAI Releases China's First Office Agent User Report: what are the key facts?

LobsterAI released China's first user behavior report for office agents, based on millions of real data points.

LobsterAI Releases China's First Office Agent User Report: what are the key facts?

Beijing leads in user volume nationwide, with overseas users accounting for over 12% and non-tier-one cities representing 62.79% of the base.

LobsterAI Releases China's First Office Agent User Report: what are the key facts?

The top 5% of heavy users consume 53.5% of all tokens, while paid users show activity levels several times higher than free users.

LobsterAI Releases China's First Office Agent User Report: what are the key facts?

Single-task scale has grown 3.1x in five months, with nearly 60% of compute consumption occurring outside traditional office hours.

LobsterAI Releases China's First Office Agent User Report: what are the key facts?

DeepSeek V4 Flash accounts for more than half of model calls, while general programming tasks lead with a 38.5% share.

Physix Frontier · News Briefing Card (enbrief · Sep 8, 2026)

LobsterAI Releases China's First Office Agent User Report

KEY FACTS

LobsterAI released China's first user behavior report for office agents, based on millions of real data points.
Beijing leads in user volume nationwide, with overseas users accounting for over 12% and non-tier-one cities representing 62.79% of the base.
The top 5% of heavy users consume 53.5% of all tokens, while paid users show activity levels several times higher than free users.
Single-task scale has grown 3.1x in five months, with nearly 60% of compute consumption occurring outside traditional office hours.
DeepSeek V4 Flash accounts for more than half of model calls, while general programming tasks lead with a 38.5% share.

KEY DATA

12.75%Overseas User Share

53.5%Top Users' Token Consumption Share

3.1xGrowth in Single-Task Scale

38.5%General Programming Task Share

PHYSIX OBSERVATION

This report marks the transition of office agents from proof-of-concept to large-scale deployment. Data reveals significant heavy usage and automation trends, as users entrust complex, long-cycle work to AI, with an emerging 'unattended' nighttime mode. This suggests that general-purpose AI productivity infrastructure is taking shape, shifting industry competition focus from raw model capability to deep integration and workflow restructuring, while the head effect indicates the market will accelerate toward high-value scenarios.

Source: enbrief original report ↗