---
title: Work Agents and the Enterprise
date: 2026-09-09
category: tech
summary: Work Agents fill the middle gap but struggle to understand the business. Without adjusting the architecture, efficiency gains are just an illusion.
draft: false
---

As domestic Work Agents like Workbuddy and Qwen Office spread in office settings, the concept of the Work Agent entered the mainstream. Actually, this situation is very split: in tier-one companies, AI long ago entered the workflow, but in companies below tier-three, getting AI into work is still stuck at the chatbot.

In March 2026 I did a survey, an internal investigation of a lifestyle-services company of around 100 people in a tier-three city (call it Company A). Over 2/3 of the people were not opposed to AI entering office workflows, and over 4/5 had already applied AI tools in their work — very similar to tier-one companies. Dig deeper and the tools Company A’s employees use are mostly Doubao (mobile more than desktop).

This survey can’t lead to any conclusion. Everyone’s business is different, the fields they’re exposed to are different too, and a lot of business simply doesn’t need AI.

But if you’ve worked in both of these two kinds of company, you’ll find that many business processes have already become two completely different modes, as different as primitive society and modern society.

I think it’s that advanced companies already have an AI Native pattern (or its prototype), while companies like Company A do not.

AI Native Team members inherently have very strong individual capability; that strong individual capability is enough to drive workflow upgrades, and they also have a good atmosphere, with more information channels than Company A. The strong get stronger, and this makes the gap between the two companies wider and wider.

Work Agents like Workbuddy and Qwen Office seized this middle gap, and they do something very unfriendly to an AI Native Team: they only try to improve efficiency inside concrete business processes.

For teams like Company A, this kind of Work Agent brings nothing short of an earth-shaking change, and it greatly frees up manpower. But as the novelty fades, the defects behind this kind of agent get exposed:

1. Completely unable to understand the company’s business
2. Context is extremely polluted, and output is poor
3. The seemingly convenient skill-marketplace feature doesn’t fit their own business
4. Token throughput is extremely slow — the speed is very slow too

I’m not writing off these kinds of agents as worthless. They still have advantages when handling a single task. Most Work Agents now offer free tokens; if you just want to make a PPT, this kind of agent’s free nature makes it a very good choice for Company A with no local compute.

Actually, enterprise AI transformation gives me a huge headache, far harder than building an AI Native Team. A company going through AI transformation simply has no way to handle its legacy business well. Transformation is equivalent to rebuilding the architecture; if positions can’t be adjusted, the architecture changes you can make are very limited. But this is the current situation most companies face: ① no redundant resources to set up a new AI Native department; ② no way to adjust the existing architecture.

If you can’t adjust the architecture, there’s no way to achieve a large efficiency gain — you can only rely on tools like Work Agent to improve efficiency. But in essence it’s still an old architecture driving a new mode, and whether it can truly improve efficiency still needs to be evaluated. This is also one of the reasons many people feel “AI is very useful, but I haven’t become less busy.”