When one person works with 16 Agents, the human begins to become the bottleneck
Gao Yan, Partner at Ruihua Smart Strategy, has configured 16 Agents for herself to handle work including product, delivery, branding, marketing, personal IP, theoretical research, strategy, and sales management respectively.
When encountering a project, she will first discuss it with the Agents, and after the solution is gradually clarified, let the Agents call tools and Skills to continue execution.
But these 16 Agents do not work at the same time. Only about 5 of them are actually kept active by Gao Yan every day.
If more than a dozen Agents are started at the same time, solutions, problems and pending confirmation items will flood in quickly. Which results are usable, which need to be modified, where to go next, and whether to continue authorization all require her judgment.
"If more than a dozen Agents work at the same time, I will soon become the decision-making bottleneck myself," she said.
Traditional management often studies how many subordinates a manager is suitable to lead.
After Agents enter the enterprise, a new question arises: how many Agents can one person manage at the same time?
This team that provides enterprise-level AI implementation services has been trying AI-native working methods since its establishment. When the workload increases, they will not immediately think of recruiting new people. Their first reaction is: can this matter be handed over to AI first? If one Agent is not enough, can AI further call other AI?
In Gao Yan's view, this is also the difference between AI-native organizations and traditional companies that only carry out "+AI" transformation.
01
AI Rewrites the "Adding Headcount" Logic of Organizational Expansion
The current organization of Ruihua Smart Strategy is very flat.
Gao Yan directly manages members of functional departments including technology, delivery, business and marketing, and each member uses their own Agents. The human organizational relationship basically only has two layers. When the team expands to dozens of people in the future, she expects that there may be one more layer, and three layers will basically be the upper limit.
As the number of people and layers increase, information transmission and collaboration costs will also rise. The same applies to Agents.
The workflow of calling one Agent to solve a problem is relatively straightforward;
If multiple Agents form an "expert group", the collaboration may slow down instead, and even fall into endless loops in mutual discussions.
Therefore, when the workload increases and the existing manpower is insufficient, the team will first judge whether AI can take on the work, and then consider adding new employees.
This working mode has been applied to specific daily work.
The company's marketing website is completed by one senior employee with the help of AI. For Gao Yan's own official account, a large number of links from topic selection, creation, typesetting to image generation are handed over to Agents.
After she confirms the topic, the Agent can continue to complete writing, modification, typesetting and matching pictures, and finally the release is completed manually due to platform permission and other reasons.
The production of PPT has also adopted a new set of methods. Gao Yan found that directly letting Agents generate traditional PPT cannot achieve ideal quality and will cause high Token cost. Later, the team switched to HTML to generate interactive presentation pages with page turning functions.
Brand specifications such as Logo, layout and copyright marks are also made into Skills.
In the early stage, the team invited part-time designers to participate in the design of Logo and basic visual specifications, and then converted these specifications into rules that Agents can call. When generating materials later, Agents will directly execute according to unified standards, and some recurring work of drawing, typesetting and PPT production in the past will be reduced accordingly.
After the business volume increases, Ruihua Smart Strategy is accustomed to adding Agents, Skills and automation capabilities first, and then deciding whether to add new employees.
This path soon encountered another problem.
02
The Faster AI Runs, the More Humans Need to Slow Down to Make Judgments
The execution speed of AI has improved, but human time has not increased.
Gao Yan cited a programming scenario: Agents can generate code all night, but the next day engineers may not be able to review all the code at all. Her own situation is similar:
When one Agent gives a solution, one judgment is needed;
If more than a dozen Agents advance at the same time, the number of nodes waiting for confirmation will increase rapidly.
Therefore, although she has configured about 16 Agents, she usually only keeps about 5 active every day. "Only in this way can I leave space for decision-making and thinking."
Five is not the standard answer, it is only her personal experience at the current stage.
Her work involves a large number of decisions and in-depth thinking. If Agents keep pushing results, the whole day will easily become a cycle of receiving, reviewing, feedback, then receiving and reviewing again. Agents can work in parallel, but the human brain has "bandwidth" limitations.
This has also changed the working methods of team members.
Everyone needs to judge by themselves which Agent to prioritize today, how to allocate time for multiple Agents, when to start new tasks, when to stop, and which results must be taken over manually.
After the Agent is configured, the task will not be automatically completed accordingly.
The goals still need to be clearly defined by humans. If the requirement is expressed vaguely, the Agent may execute along the wrong direction, generate a large number of lengthy and inaccurate results, and increase Token consumption at the same time.
Therefore, Gao Yan regards goal decomposition, task allocation and result review as the basic capabilities that members need to possess. These tasks that used to be closer to the work of managers now can also be handled by people without management titles who dispatch multiple Agents every day.
Traditional management has the concept of "span of management", which discusses how many subordinates a manager is suitable to lead. Gao Yan believes that similar problems will arise in the AI era, that is, how many Agents can one person manage at the same time?
And the number is not the more the better. When Agents keep working in parallel, they will push more and more results and decision-making nodes to humans.
Human attention is constantly divided, and review and judgment will soon fail to keep up with the output speed of machines. Gao Yan also mentioned that staying in this state for a long time will affect people's work well-being and sense of achievement.
Agents can be increased continuously, but human judgment ability cannot be expanded synchronously.
03
Posts Are Reduced and Redesigned
After Agents are introduced into work, some posts have started to shrink.
The first type of posts Gao Yan mentioned are middle managers who used to only be responsible for transferring information upwards and downwards. The work of this group is generally to collect data from lower levels, organize it and then report to superiors.
If information can be directly organized and transmitted by the system, the space for this kind of work will be significantly reduced. The other type is pure tool-based posts. Simple drawing, typesetting and PPT production can already be largely handed over to Agents.
At the same time, some work has become more demanding. FDE is one of them.
Ruihua Smart Strategy provides To B services. Gao Yan found that deploying Agents (which she calls "silicon-based employees" on duty) to client enterprises is only the beginning. Even if many enterprises get Agents, they may not know how to use them safely, efficiently and at low cost; the original processes, permission systems, post division of labor and personnel capabilities will not change automatically with the launch of AI.
Therefore, they have expanded the scope of FDE's work.
Neuters
If the client has low AI maturity, FDE may need to lead the expert team to provide long-term guidance;
If the client has high maturity, the service will gradually become lighter. They pay more attention to whether the client has really used the technology and whether it has generated business value.
The person who is in charge of FDE now used to work in digital marketing. He was not recruited from the market as a "mature FDE".
The team valued his technical research ability, professional orientation and service awareness, and first let him participate in internal AI transformation and client delivery, then gradually take the role of FDE lead.
Gao Yan believes that when a new post first appears, it is difficult to find mature talents directly in the market. The team will first look at what problems this work needs to solve and what capabilities are required, then select people with corresponding traits from existing members, and cultivate them gradually in practical work.
Knowledge management has also been repositioned as a key priority. Even with a very small team, Ruihua Smart Strategy has specially set up the knowledge management function.
Gao Yan has a knowledge management Agent. In addition to organizing past precipitated content into a structured knowledge service layer, it will also regularly "distill" long-term valuable ideas from daily interactions, content, speeches and materials, and continuously enrich this knowledge service layer for all other Agents to access and call.
In this way, the Agent in charge of personal IP will be more and more familiar with her expression habits, and other Agents can also inherit the methods and judgments that have been precipitated.
But she does not advocate stuffing everything into the knowledge base.
There is no need to save all the chat content, private information and low-value content in IM; the cost of full cleaning of historical data may also be higher than the benefit. She prefers to keep the knowledge that can survive across cycles.
In the past, knowledge management mainly solved the problem of "whether the content can be found later". Now the precipitated experience, rules and methods will continue to be imported into Agents to affect future work.
04
80% of Employees Are Senior Veterans, but Weekly Reports Are Abolished
Ruihua Smart Strategy has another feature that does not seem very "AI-native": about 80% of the team members have more than 15 years of work experience.
Gao Yan believes that this is related to To B services.
Clients will not become AI-native enterprises overnight just because AI appears. The service team needs to understand how the client's original organization operates, how the business process runs, which links have high risks, which places can be improved, and how to design a transition path that the client can accept.
These things rely on long-term accumulation. So at this stage, they first choose people who are rich in experience and can use AI deeply at the same time.
But experience can also form inertia.
Gao Yan, a post-70s generation herself, also constantly reminds herself that the working methods formed in the past may become constraints. In the future, the team will introduce more young members, including post-2000s generation.
Young members are often more likely to directly use new AI tools and infrastructures, without having to abandon the working methods they have been familiar with for many years first.
Both types of people have their own value: one side has industry Know-how, and the other has the "native" sense of AI.
In terms of daily management, Ruihua Smart Strategy has made a lot of simplifications.
The team has no fixed morning meeting and no weekly report. They determine the strategy, division of labor and milestones to be completed at the end of the quarter at the beginning of the quarter, and hold meetings only when difficulties arise, progress needs to be synchronized or key information needs to be shared.
Gao Yan does not agree with relying on high-frequency reports to maintain control.
She pays more attention to whether the common cognition, values and boundaries are clear. Of course, this method is also related to the size of the team. After the number of people continues to increase, many problems that can be solved through direct communication today may still need mechanisms and processes again.
The increase in the number of people will make the collaboration network more complex, and the same is true for the increase of Agents.
At the end of the interview, Gao Yan talked about a more fundamental question: if execution, retrieval, drawing, writing and programming are increasingly handed over to AI, what is left for humans to do?
She gave two answers.
One is responsibility. AI can provide solutions and participate in decision-making, but it cannot bear the consequences. In enterprise-level scenarios, where to authorize, which result to adopt, and who is responsible for taking remedial measures when problems arise, ultimately fall on humans.
The other is the connection between people. Especially for large client sales, consulting and services, trust is difficult to be directly established by a complete solution generated by an Agent. A word on site, a feedback, an understanding, the unique internal atmosphere and "political issues" that people perceive when they arrive at the client site may all affect whether the cooperation will continue.
Gao Yan only keeps about 5 Agents truly active every day. It seems to be just a usage habit, but it has already touched a new constraint after AI enters the organization.
In the past, companies worried that no one would work. Now, as Agents become more and more capable, people begin to worry that they cannot keep up with the review work.
After the execution capability increases rapidly, judgment is becoming a new scarce capability.
This article is from WeChat official account "Neuters" (ID: Neuters), author: Alex, authorized by 36Kr to release.