Four AI Collaboration Modes to Safeguard Managers' Core Value

Which work links still rely on human subjective initiative, and which can be safely handed over to AI to complete?

I have a client, let's call her Susan for the sake of privacy, who is the CEO of a media company. The firm has just gone through a tough fiscal quarter, and she was preparing for an all-hands communication meeting. The office director used AI, referencing her past speech manuscripts and internal performance memos, to draft the speech. The draft was logically coherent, appropriately worded, and aligned with the company's development priorities. But when Susan reviewed it, she always felt something was off: the speech was overly conservative. It did not face up to the real situation at the moment, but only patched together the reassuring remarks she had repeatedly conveyed in the past.

This experience reflects a more widespread dilemma. Over the past year, Susan has integrated AI into the daily workflow of the executive team. The initial results were very impressive: customer insight analysis that used to take several days to complete can now be produced in a few hours; strategic documents can integrate a large number of viewpoints from conversations; analytical work is faster and covers a wider range. However, this gain comes at a cost. The more the team relies on AI, the more convergent their way of thinking becomes. The outputs produced by everyone increasingly tend to choose the most secure and non-controversial viewpoints — exquisitely phrased and fully demonstrated, but completely lack distinctive features.

AI is extremely good at reinforcing people's inherent mindsets. As a tool, it is well suited for spreading existing knowledge, but it is difficult to guide people to question inherent assumptions, deal with uncertainty, and explore new directions. Such scenarios cannot be separated from human initiative. More and more managers will face the dilemma encountered by Susan: which work links still rely on human subjective initiative, and which can be safely handed over to AI?

The Critical Role of Human Initiative

In the workplace context, human initiative refers to the ability to actively make choices and put them into action: exercising judgment, taking a clear stance, and maintaining workplace interpersonal relationships. In the process of providing consulting for executives, I found that managers' human initiative is mainly reflected in three dimensions: judgment, expressive discourse power, and in-person presence influence. The combination of the three can help leaders cope with key scenarios that require taking responsibility and winning trust. In these fields, AI can provide information support and assist decision-making, but cannot take full responsibility.

After observing many executives who embrace AI, I found that seemingly trivial handovers of work will gradually erode managers' human initiative. Leaders become more and more passive in communication, and neither themselves nor others can clearly see how much value their own judgment, verbal expression, and personal participation can create. In the long run, managers' contributions are weakened, and their personal leadership image is also damaged.

To preserve human initiative, the first step is to distinguish: which links must rely on your judgment, personal expression or personal participation to ensure the quality of the final output. The top executives I have worked with adopt a set of methods: divide four AI collaboration modes according to the tasks at hand. In the first three modes, the unique value of human beings must be at the core; in the fourth mode, low-risk work is handed over to AI to free up energy for in-depth thinking.

Practical Methods for the Four Collaboration Modes

You can use this framework to preserve your own human initiative, and the following content is attached with real cases provided by the executives I have served.

1. Leader-Led Mode

Preserve your judgment: make independent judgments first, and then refer to AI's opinions

The work that requires leader leadership relies corely on your judgment — when data alone is not enough to diagnose and solve business problems, it is up to you to define what is most important. It also includes weighing the trade-off solutions acceptable to the organization, and determining the risk tolerance level on the premise that all choices have costs.

AI can only play an auxiliary role: organizing materials and comparing alternative solutions. But it cannot clarify the historical background, internal games, interpersonal conditions, objective constraints and potential consequences that leaders must deal with and explain to all parties. Therefore, the right to diagnose problems and define problems initially must be in your hands. The way the problem is defined will affect all subsequent solutions and determine which countermeasures can be truly implemented.

I once served the division president of an information and data enterprise. Every quarter, he leads the team to review the profit margin and sort out the driving factors of business profitability. He began to use AI to summarize data in advance, so that he could get familiar with various indicators before the meeting, instead of studying them on the spot temporarily.

Over time, he noticed a problem: the content of AI's summary directly determines the focus of the meeting discussion. Whatever the abstract highlights, the team will interpret all information from this perspective. In one quarter, the profit of the business division continued to decline, and the team traced the root cause. The AI summary listed the rise in supplier costs as the core factor at the beginning. Most of the whole meeting was spent discussing this reason, until later in-depth data mining found that the real core incentive was that the sales channel tilted to distributors, who kept striving for greater discounts and suppressed the profit of a single order. Because a lot of time was spent discussing supplier costs in the early stage, the team took a long time to find the real crux and conceive solutions.

This client adjusted his work habit: before checking the AI's data summary, he first wrote down his judgment and basis for the core issue, and then used AI to verify his views. Sometimes AI confirms his judgment, and sometimes it raises doubts. No matter what the result is, he is more confident when attending the meeting, can firmly output his own views, and respond to various questions more calmly.

Practical suggestions for leader-led work:

Before using AI to analyze business problems, form independent judgments first.

Write down the real problem you identified, decision-making criteria or preliminary inferences, even if it is only short and rough text.

Then let AI question your point of view, provide a new perspective, or point out the information you missed.

This process can continuously exercise managers' decision-making judgment. Once skipped, this ability will degrade rapidly.

2. Leader-Shaping Mode

Preserve your discourse power: communicate in the language that you will actually use and are willing to be responsible for

Leadership communication conveys more than just information. Depending on the scenario, it can convey calmness, empathy or firm belief. Subordinates will pay attention to: whether you see the challenges clearly, understand the impact brought by the challenges, and whether everyone can trust you to lead the team through difficulties.

This is why personal unique expression is so important. You have the ability to face the cruel reality directly, release unpopular news, and clearly show your stance, so that the text can carry your original intention, standards and humanistic temperature. All kinds of communication manuscripts in leadership shaping related work — speeches, board materials, crisis notices, organizational adjustment announcements, and all texts used to point out new directions, are inseparable from this quality. AI can optimize the structure, clarity and logic of the manuscript, and even provide alternative wording. But the core conception, tone and final expression must be finalized by you.

It is easier said than done. If the first draft is directly generated by AI, it is difficult to retain personal expression. Even if you modify the manuscript later, the initial framework, language and preset ideas built by AI will still subtly affect the final draft. The final manuscript may be more organized than the version written by yourself, but it lacks a sense of reality. In fact, a study on large model assisted writing shows that people who rely heavily on the model to write manuscripts have the lowest sense of belonging to the finished products. This is worth alerting all managers: your speech needs to come from your heart; when the board of directors, investors, and the media raise questions, you must stand firm and always adhere to the views and expressions you conveyed earlier.

Susan's case is the best proof. Back to the incident of the all-hands communication meeting speech, she adjusted the way the team uses AI. Instead of letting AI write the full draft directly, AI generates the key point outline, and Susan expands independently on this basis. After completing the first draft, she read the manuscript aloud and immediately screened out two types of content: sentences that she would never say in daily communication, and expressions that do not match the current severe situation. Most of the text was rewritten by herself. The team commented that this was the most transparent and most emotional communication she had ever delivered.

Practical suggestions for leader-shaping work:

Clarify the original intention of communication first. After the audience reads and hears the content, what do you want them to understand, what feelings do you want them to have, and what actions do you want them to take. Sort out three to five core points to convey. These viewpoints come from your judgment of the situation and cannot be handed over to AI to generate.

Let AI sort out, improve and review these key points. For reference: "Write an outline for the all-hands communication meeting, face up to the tough fiscal quarter, take the initiative to take responsibility, explain the adjustment plan, and help everyone see the follow-up direction clearly."

Use the AI outline as the framework to independently complete the first draft. Read through the manuscript, mark all two types of text: sentences derived from the outline that you will not say verbally, overly templated sentences, and paragraphs that weaken the severity of the situation, and rewrite them in your own language.

As AI's communication and reasoning capabilities continue to enhance, the value of creating exclusive personal expression will become more prominent. Anyone can use tools to output viewpoints, but using tools to amplify unique personal expression will become a scarce and extremely valuable ability. This uniqueness is the key to making you stand out.

3. AI-Assisted Mode

Preserve your in-person presence influence: maintain direct contact with people and do not be isolated by AI

In-person presence influence means getting close to team members and capturing details that AI reports cannot present: subtle changes in tone, casual ideas, and signals that can only be detected by being present in person. This is the most difficult of the three capabilities to preserve. Managers receive a large number of streamlined materials every day: summaries of customer feedback, meeting minutes, summaries of candidate interviews, key points of employee opinions, and lists of alternative solutions. AI does not weaken the value of such materials, but makes the summary content easier to produce and easier to read. But no matter how it is generated, the abstract can only show the general outline of things.

This brings two major risks:

First, over-reliance. The continuous stream of seemingly logically reasonable streamlined materials will make the review a mere formality. Managers who used to read carefully and dig deeply now just browse quickly; those who used to check carefully now easily approve. The closer the review is to a simple stamp of approval, the more managers are separated from the real people and information behind the abstracts.

Second, information estrangement. The summary materials will make people have the illusion that they have fully understood the situation. In fact, you have been separated from the real front-line scene.

I once served a regional CEO who relied on AI-generated information summaries to follow up on customer meetings and employee feedback across a large area. The summaries saved a lot of time and were very practical. He mastered all the factual information, but gradually felt increasingly distant from the core business personnel. The summaries cannot capture those subtle information: the hesitation of customers when facing thorny problems, the ideas that employees privately reveal after the meeting, and the hidden dangers that are only mentioned once but are more important than all the items on the agenda. Without these clues, even though he has always been known for being pioneering and good at business growth, he has become more hesitant in major decisions and unwilling to actively explore new opportunities.

He did not abandon the AI summaries, but re-established direct communication. He re-arranged customer visits, one-on-one communications, and open-ended exchanges with the team without preset scripts into his work schedule. These interactions helped him re-establish business intuition, restore decision-making confidence, and be more willing to dig into the motivations behind performance and actively explore new directions for business development.

Practical suggestions for AI-assisted work:

Before relying on AI to summarize content, formulate review rules in advance: which content needs secondary manual review, which needs to trace the original materials, and which needs to be focused on checking for missing multi-party voices and hidden preset biases.

Develop the habit of asking AI questions: What might this summary have missed? Don't just judge whether the content is reasonable. A fluent and determined output is not equal to complete and objective information.

Ensure unscripted direct communication. Even if the summary can cover relevant information, still arrange face-to-face communication, including customer visits, one-on-one conversations, and informal team communication.

There are two levels of goals: continue to rely on people to interpret information, and at the same time let managers directly connect with customers, employees and stakeholders behind each summary.

4. AI Full-Processing Mode

Free up energy: hand over all work that does not require your judgment, personal expression or personal participation to AI

The fourth mode makes room for the other three types of important work. AI full-processing work is low-risk standardized tasks that require little or no supervision: schedule arrangement, demand diversion, regular report typesetting, routine question response, and standardized first draft writing. Stripping such tasks from the manager's work list allows you to focus on the work that only you can complete.

The popularization of agent technology makes the definition of boundaries more critical. Agents can connect multiple operations, call other tools, transmit information across programs, and continue to advance tasks without waiting for human instructions at every step. This autonomy is the advantage of agents, but it also means that work boundaries must be clearly defined. A chief operating officer I cooperated with will definitely ask four questions before enabling AI to handle tasks fully:

1. Can this work content be clearly defined?

2. Can the team clearly judge the quality of the results?

3. Is it easy to detect errors and the correction cost is low when mistakes occur?

4. Is it a routine internal work, or a work where AI represents the corporate image externally?

For example, demand diversion and first draft writing of conventional product consultations all meet the boundary requirements. However, writing an apology letter to important customers after a service failure is not in this category, as the tone of the information directly affects the overall customer cooperation relationship.

Remember: such boundaries must be set actively. For each task handled by the agent, make it clear in advance: the scope of execution without asking for instructions, what situation requires manual intervention and verification, and who is responsible for the problem after it occurs. At the same time, convert the operations that originally relied on intuitive judgment (when to escalate problems, when to take the initiative to intervene) into clear rules for the agent to execute.

Practical suggestions for AI full-processing work:

Use the four questions above to filter tasks that are suitable for AI to handle.

Publicly clarify the scope of work adapted to this mode, and strictly keep the boundaries.

Ensure that the team is clear about AI's autonomous execution authority, manual intervention nodes, and result responsibility ownership.

Whether this mode can exert value depends on whether the boundary rules are clear and can be continuously implemented.

Managers who can gain the most value from AI will actively plan their own work. Most importantly, they know which scenarios to exert their judgment, personal expression and in-person presence influence: the critical moments of defining problems, conveying key information, and maintaining trust and team cohesion.

AI can make managers more efficient, more informed and more capable. But the same tool can also give rise to templated work results, make people's perfunctory reviews, and separate them from the real front-line business. If managers want to preserve their own human initiative, they must continuously and actively clarify: what AI can undertake, which content must be manually reviewed, and which work can only be completed by themselves in person.

David Lancefield | Article

David Lancefield is a corporate transformation consultant, strategy expert, and executive coach. He has provided consulting for more than 50 CEOs and hundreds of executives.

Zhou Qiang | Proofreading

This article is from the WeChat official account "Harvard Business Review" (ID: hbrchinese), written by HBR-China, and republished with authorization from 36Kr.