Are AI cloning offices really more efficient?

Last Friday, my friend Lao Wang's startup launched a new system: each employee can clone an AI version of himself with one click, process emails, write plans, and follow up with customers. During the resumption of trading on Monday, Lao Wang discovered a strange phenomenon-the company's overall output fell instead of rising, but the meeting time increased by 30%. What's even weirder is that employees began to question each other: "Did you write this reply yourself or is it your AI agent? "

This is not an exception. In the past six months, I have followed seven companies piloting AI office agents and found a common rule: When the number of AI agents exceeds 30% of the team's number, the overall efficiency begins to decline . Not because AI is not smart enough, but because collaboration costs are snowballing.

Invisible synergy tax

The core selling point of AI agents is "releasing human time." Products like Munde Difflin claim that an AI agent can do 60-80% of repetitive work. But what they don't tell you is that the remaining 20-40% is not simply a "manual finishing"-it's an exponential increase in collaboration costs.

For example: After a 50-person marketing company introduced an AI agent, the average number of messages received per employee per day increased from 82 to 147. Why? Because AI agencies:

1. automatically follows up with customers (generates new messages)
2. replies to colleagues 'AI agents (dialogue between AI)
3. requires human confirmation (added confirmation link)

To make matters worse, not all of these messages are valid information. A product manager I interviewed told me: "I used to only need to read 10 emails, but now I have to sift through 30 AI-generated emails to determine which ones really need me to deal with. This is more tiring than writing your own email. "

dominoes that collapse trust

The second problem caused by AI agents is the collapse of trust. When you are not sure whether the other person is a human or an AI, all communication requires additional confirmation.

I know a startup that does SaaS that is trying to use AI agents to handle customer support. It was found that:

• customer satisfaction dropped by 18%(because AI answers were too "official")
• manual support workload increased by 40%(due to the need to correct AI errors) The - team began to suspect each other: "Did you write this reply yourself? "

The most ironic thing is that in order to solve this problem, the company had to introduce a "trust scoring system" that allowed employees to rate responses from AI agents. The result is that employees spend more time scoring than answering customers directly.

This reminds me of the "Prisoner's Dilemma" in economics: When everyone pursues maximizing individual efficiency, overall efficiency declines . AI agents make each individual seem more efficient, but the entire system is caught in a more complex game.

Lao Zhang's Data Center

I know an operation and maintenance engineer Lao Zhang. His company launched an AI operation and maintenance agent last year. Each agent can automatically handle alarms, optimize resources, and even predict failures. In theory, this should greatly reduce Lao Zhang's workload.

But in fact, Lao Zhang's work has become more complicated. Every morning, he spends an hour checking the AI agent's decision log to make sure there are no misoperations. Once, an AI agent mistakenly marked a critical server as "idle," causing a service outage for 3 hours. A later review found that the error was due to another AI agent providing wrong data.

Lao Zhang told me: "Before, I only had to deal with real failures. Now I have to spend 60% of my time confirming that the AI agent is doing the right job. Sometimes I'd rather do it myself, at least I know what I'm doing. "

This story reminds me of a counter-intuitive rule: When the tool becomes powerful enough, supervising the tool itself can be more time-consuming than using the tool . An AI agent is not a simple "assistant", but an "employee" who needs to be managed.

Opposition: Is this just a transition pain?

Some people will say that this is only a problem in the early stages of the development of AI agents. As technology advances, these problems will be solved naturally. For example:

1. AI will become smarter and reduce errors
2. collaboration mechanism will be optimized to reduce redundant communication
3. People will gradually adapt to the existence of AI agents

This view makes sense, but it ignores a key issue: Collaboration costs are growing much faster than AI capabilities are improving .

For example, after the release of GPT-4, AI's writing capabilities have indeed made a qualitative leap. But what followed was:

• mail volume increased by 40%
• meeting time increased by 25%
• confirmation link increased by 50%

This shows that the improvement of AI capabilities cannot be linearly transformed into an improvement in overall efficiency . Because while AI capabilities are improved, it also brings more "creative work"-such as judging whether AI's decisions are correct, coordinating the work of multiple AI agents, handling conflicts between AI agents, etc.

QKPFX16 The Cost of QK Efficiencism

The rise of AI agents is essentially another victory for efficiency. We always pursue doing more in less time, but ignore a basic fact: Efficiency is not free .

Every efficiency improvement is accompanied by new costs:

• Industrial Revolution improves production efficiency, but brings environmental pollution
• Internet improves the efficiency of information dissemination, but brings information overload
• AI agents improve individual work efficiency, but bring collaboration costs

This reminds me of the "second law" of economics: When you try to optimize a certain part of a system, the efficiency of the entire system may decrease . AI agents are a typical example-they optimize individual work efficiency, but the efficiency of the overall system decreases.

What do we really need?

AI agents are not a scourge, but they are not a panacea either. The real question is: Do we really need this "efficiency"? *

I interviewed a senior HR person, and she told me an interesting phenomenon: those companies that first introduced AI agents had a higher employee turnover rate. Why? Because employees feel that their jobs have been replaced by AI and lose their sense of accomplishment.

This made me think that maybe what we really need is not more efficiency, but more meaningful work. AI agents can handle repetitive work, but they cannot replace real connections between people, creative thinking, and enthusiasm for work.

Lao Wang's company ultimately decided to limit the scope of use of AI agents: only to handle standardized customer inquiries, while reserving all creative work for humans. It turned out that the company's overall efficiency increased by 15%.

This incident made me realize: Efficiency is not an end, but a means . When we sacrifice the meaning of our work and the connection between people for efficiency, we may have deviated from our original intention.

's final thoughts

The story of the AI Clone Office continues, but one thing is already clear: Increased efficiency does not always lead to better results . Sometimes, less AI may mean more humanity, and less collaboration costs may mean higher overall efficiency.

I make a bet: In the next two years, more and more companies will realize this and start to limit the scope of use of AI agents. Not because the technology is not advanced enough, but because we have finally realized that some jobs are more suitable for people to do.

At the end of the day, technology is just a tool, and the value of a tool depends on how we use it. Instead of pursuing unlimited efficiency, it's better to ask yourself: What kind of work and life do we really want?