ºÎ»ê½Ãû µµ¼­¿ä¾à
±¹³»µµ¼­ ¿ä¾à
±¹³»µµ¼­ ÇÁ¸®ºä ÇØ¿Üµµ¼­ ÇÁ¸®ºä
±Û·Î¹ú Æ®·»µå ¹Ìµð¾î ºê¸®Çνº
   ¹Ìµð¾î ºê¸®Çνº 

åǥÁö





  • How to Use AI as a Team Tool, Not Just a Personal Tool


    AI has rapidly established itself as a personal assistant that writes reports and summarizes materials. However, when AI participates in meetings and collaboration, it changes who gets to speak, how discussions unfold, and who takes the lead in making judgments. To improve team performance, organizations must move beyond learning how to use AI and design rules for how humans and AI should work together.

    [Key Message]
    * Strong individual AI skills do not automatically lead to better team decisions. Team performance depends not only on speed, but also on participation, coordination, trust, and accountability.

    * Prompts should be designed collectively by the team. If one person controls how questions are framed, that individual¡¯s perspective may quietly become the team¡¯s definition of the problem.

    * AI should serve in multiple roles, such as critic, customer, competitor, future observer, and facilitator, rather than acting as an authority that provides the correct answer. These roles challenge familiar assumptions and broaden the scope of discussion.

    * AI-generated responses should be treated as material for verification, not as meeting conclusions. Teams must distinguish facts from estimates, assumptions, and opinions, and cross-check important claims against original sources, experts, customers, and frontline information.

    * Final judgment and accountability must remain with people and the team. Organizations should document not only what AI proposed, but also what the team selected, why it made that decision, and who is responsible for the outcome.

    ***

    Individual Productivity Does Not Guarantee Team Performance
    Generative AI is rapidly changing the way individuals work. It drafts emails, extracts key points from lengthy reports, organizes ideas, and turns complex data into easy-to-understand sentences. As more people experience tasks that once took hours being completed in minutes, AI has become an everyday workplace tool rather than an exceptional technological innovation. Companies are also providing employees with AI accounts, offering training programs, and encouraging its use.

    However, using AI well as an individual and using AI well as a team are entirely different matters. In individual work, only the relationship between the user and AI needs to be managed. If the desired result does not appear, the user can revise the question or ask again using different wording. The individual can also decide whether to accept the AI¡¯s response. A team meeting, however, brings together people with different roles, areas of expertise, interests, and levels of authority. A response that is useful to one person may be a proposal that overlooks important conditions for another. An efficient summary may feel to someone else as though their opinion has been erased.

    Individual productivity is fundamentally a matter of work speed and quality. Team performance adds other elements, including coordination, participation, trust, and responsibility. Even if every team member uses AI to produce more material, it does not necessarily mean the team will make better decisions. If individuals converse with different AI systems and produce materials based on different assumptions, the amount of information may increase while shared understanding weakens. The documents submitted to meetings may become richer, but it may also take more time to determine which materials should be trusted and how conflicting analyses should be reconciled.

    This shift becomes increasingly important as the focus of AI use moves from individuals to teams. A Capgemini Research Institute survey of 500 executives worldwide projected that the proportion of organizations actively using AI in team meetings would more than triple within the next three years. Corporate expectations are also high. Companies believe that if AI can summarize meetings, provide information immediately, and suggest a variety of alternatives, teams may be able to reduce meeting time while achieving better outcomes.

    The problem is that adding AI to meetings does not automatically improve collaboration. If existing meetings are dominated by a small number of voices, information is not adequately shared across departments, and responsibility remains unclear even after decisions are made, AI may amplify these problems more quickly rather than solve them. When powerful technology is placed on top of an imperfect collaborative structure, it can reinforce the structure¡¯s existing flaws.

    Therefore, the first question to ask when introducing AI at the team level is not, ¡°Which AI should we use?¡± Teams should instead ask, ¡°How will we formulate questions, whose perspectives will we include, how will we verify AI¡¯s suggestions, and who will be accountable for the final judgment?¡± This is why the principles of collaboration must be established before the technology is selected.

    Unexpected Problems That Arise When AI Enters the Meeting Room
    The simplest way to use AI in a meeting is for one person to open a computer and enter questions on behalf of the team. On the surface, this appears efficient. Participants express their opinions, the designated person writes the prompt, and AI produces a response within seconds. Yet this approach creates a subtle shift in power. The person operating the AI chooses the wording and scope of the question, as well as which conditions to include and which to exclude.

    A prompt is not merely a command. It is a small decision-making structure that determines what will be treated as the problem. The question ¡°Suggest a sales strategy for our new product¡± will produce a different response from ¡°Suggest a way to increase sales of our new product while reducing the loss of existing customers.¡± The result also changes depending on whether conditions such as cost, brand value, field personnel, and technical constraints are included. If one person manages this entire process, that person¡¯s perspective may unconsciously become the team¡¯s collective question.

    Other team members may gradually be pushed into spectator mode. When an AI-generated response begins appearing rapidly on the screen, people may stop talking and start reading the result. Previously, they developed ideas by questioning and challenging one another, but they may now be confined to evaluating sentences produced by AI. The meeting shifts from a conversation between people to a relationship between people and a screen. Team members with less seniority or fewer opportunities to speak may encounter a polished AI response before they have had enough time to explain their own ideas.

    Discussions may also become fragmented. If participants enter whatever questions come to mind into AI as the meeting proceeds, the number of responses increases, but those responses may not connect into a coherent line of thought. The team may ask about market size, move on to competitor analysis, immediately generate promotional copy, and then ask about potential risks. Each response may sound plausible, but it may be based on a different set of assumptions. The team feels that it has gathered a great deal of information, yet what it actually needs to decide becomes even less clear.

    Teams should also be wary of the first response generated by AI becoming the reference point for the entire discussion. People tend to adjust later judgments around the first number or explanation they encounter. If AI evaluates a particular market as promising early in a meeting, the subsequent discussion is likely to shift from whether the company should enter that market to how it should enter. A proposal that has not yet been verified can harden into a working assumption rather than simply serving as a starting point for discussion.

    The fluency of AI-generated language makes judgment more difficult as well. AI can present uncertain information in logical and confident language. Even when the evidence is weak or important context is missing, participants may accept the output as high-quality analysis because it appears complete and polished. In time-pressured meetings, people are even more likely to say, ¡°The general direction seems right,¡± and proceed to the next step instead of checking every source and condition.

    The most serious problem is that ownership of the judgment becomes blurred. If a team selects an AI-generated proposal and the outcome is poor, it may be unclear whose judgment it was. The person who entered the question, the participants who agreed with the response, and the leader who chaired the meeting may all shift responsibility to one another. The statement ¡°That is what the AI concluded¡± may sound like an explanation, but it actually reveals a gap in accountability. AI can generate a proposal, but it cannot take responsibility for the consequences of adopting that proposal.

    When AI lowers the quality of a meeting, the cause lies less in the technology itself than in an unprepared method of using it. If a team activates the tool before deciding what roles members will play and at which stages AI should intervene, participation may decline, the discussion may fragment, and control over judgment may move outside the team. Meeting time may become shorter, but the time devoted to careful consideration may shrink as well.

    The Entire Team Should Engage in the Conversation with AI
    The first principle of using AI as a team tool is to make interaction with AI a collaborative activity. Teams must move away from a model in which one person asks questions on behalf of everyone and simply delivers the answers. Members should participate in the process of formulating the questions themselves. A good prompt is not a technically sophisticated sentence. It is a question that faithfully incorporates the team¡¯s diverse knowledge and concerns.

    At the beginning of a meeting, the team should agree on the scope of the problem before activating AI. Members should first discuss what problem must be solved, what outcome is needed, what constraints must be considered, and which assumptions have not yet been verified. Sales personnel may present customer reactions, finance personnel may address costs and profitability, technical personnel may discuss feasibility, and field personnel may identify operational risks. When these perspectives are reflected in a single question, the AI¡¯s response becomes more relevant to the team¡¯s actual circumstances.

    The team also needs to explain its background to the AI. Providing information about members¡¯ roles and expertise, the project environment, approaches already attempted, and reasons for previous failures can produce more specific analysis rather than generic advice. However, this information may include customer data, trade secrets, or personal information, so the company¡¯s security policies and data-use standards must be followed. Entering more information is not always better. Teams should provide the necessary context while clearly defining which information must never be entered.

    Instead of having one person complete the prompt and then show it to everyone, the prompt can be displayed on the screen and revised collaboratively. Team members may point out that the customer perspective is missing, that budget constraints should be included, or that a particular expression already assumes a certain conclusion. This process may appear somewhat slow, but it is an important stage in aligning the team¡¯s thinking. As members formulate the question together, they understand what others consider important and develop a shared language for discussing the problem.

    Even after AI produces a response, every member should participate in revising and challenging it. One effective method is to examine, in turn, which parts each person agrees with, which parts differ from reality, and what information has been omitted. The AI response should be treated as material to be reviewed, not as the conclusion of the meeting. At this point, the leader should hear other members¡¯ opinions before offering an evaluation. If the leader reacts positively to the AI response first, members with opposing views may remain silent.

    The team also needs roles for managing its interactions with AI. However, if one person remains responsible for prompting, authority may once again become concentrated. Teams can therefore rotate the role from meeting to meeting or divide it by stage. One person might enter the questions, another verify the evidence behind the responses, and another identify missing perspectives and counterarguments. Dividing these roles also reduces the risk that the whole team will become dependent on the skills of a single person who is proficient at using AI.

    Mechanisms can also be introduced to maintain a balance of participation. Before using AI, team members might independently write down their opinions. After seeing the response, every participant might be required to voice one concern. Securing independent viewpoints first can reduce the tendency to be overly influenced by the AI¡¯s initial response. It also makes it more likely that the question will include not only the views of the loudest people but also those of people who understand field operations, work closely with customers, or possess different experiences.

    What matters here is not merely that everyone is looking at the same AI screen. It is the sense that team members are participating together in shaping the question and interpreting the response. When AI is regarded as a shared resource, members do not passively consume its output. They actively intervene in the process. The essence of team-based AI does not lie in sharing a single account. It lies in sharing authority over questioning, verification, interpretation, and judgment.

    Give AI Multiple Roles Instead of Just One
    AI¡¯s role in meetings usually begins with recording and summarization. Its ability to organize meeting discussions, extract action items, and document decisions is useful. However, using AI only as a note-taker means missing its broader potential to improve collaboration. AI should not be treated as an authority that makes decisions on behalf of the team. It can instead take on multiple roles that expand the range of the team¡¯s thinking.

    One representative role is that of a critic. When the team quickly agrees on a particular course of action, it can ask AI to identify why the decision might fail. The team can ask which assumptions are most vulnerable, what objections might be raised from an opposing position, and what unexpected side effects could arise. This role is especially important in teams where internal agreement is strong. Members may feel a relational burden when raising objections directly, but presenting an opposing argument through AI can open the door to discussion more comfortably.

    AI can also take on the role of a customer. When considering a new product or service, a company¡¯s internal logic can easily remain confined to the supplier¡¯s perspective. By asking AI to raise questions from the viewpoints of a price-sensitive customer, a customer unfamiliar with digital environments, a customer satisfied with an existing product, or a customer who has experienced dissatisfaction, the team may discover friction points it had overlooked. Of course, AI cannot perfectly represent actual customers. It should be used to identify questions that require further investigation rather than to replace customer research and field interviews.

    AI can also assume the role of a competitor or external stakeholder. It can examine how competitors might respond to the company¡¯s strategy, what risks regulators might identify, and what burdens the strategy could impose on business partners. Reviewing the same agenda repeatedly from different positions can reveal effects that were invisible when viewed only through the team¡¯s internal interests. Strategic decisions that appear reasonable within the organization may produce different outcomes depending on the response of the wider market, so external perspectives need to be introduced deliberately.

    The role of a future observer can also be meaningful. The team can ask AI to look back on today¡¯s decision from one or three years in the future and consider what it might regret or which risks that seem small today might grow over time. Meetings focused on short-term results can easily overlook the long-term loss of capabilities, customer trust, or changes in organizational culture. Shifting the point in time through AI can broaden the range of judgment.

    When a meeting becomes stagnant, AI can serve as a facilitator. It can be asked to summarize the similarities and differences among the opinions expressed, identify questions that remain unanswered, and suggest options for advancing the discussion. However, if AI organizes the debate too smoothly, it may appear as though genuine conflict has disappeared. If there are fundamentally conflicting interests hidden beneath different forms of expression, those conflicts must be made explicit. A good summary does not conceal disagreement. It distinguishes what has been agreed upon from what remains unresolved.

    The purpose of assigning AI multiple roles is not to receive more answers. It is to examine a single problem from several angles and challenge assumptions that the team has taken for granted. Questions such as ¡°Present three ways this strategy might fail,¡± ¡°What concerns would different types of customers have?¡± and ¡°What conditions would make the opposite conclusion valid?¡± deepen the team¡¯s thinking more effectively than asking, ¡°Tell us the best strategy.¡±

    The moment a team asks AI for the correct answer, AI can easily appear to be an authority. By contrast, when the team requests counterarguments, questions, scenarios, and alternatives, AI becomes a collaborator that supports thought. Teams do not always need a rapid conclusion. Sometimes it is more important to delay premature agreement and discover possibilities that have not yet been examined. AI¡¯s true value is revealed less in its ability to produce answers quickly than in its ability to encourage teams to formulate better questions.

    Judgment and Responsibility Must Remain with the Team
    The final principle of using AI as a team tool is to define clear boundaries around judgment and responsibility. AI can analyze materials, generate options, and compare arguments, but it cannot independently determine an organization¡¯s goals, values, or tolerance for risk. Even when presented with the same information, different choices may be necessary depending on the company¡¯s circumstances and responsibilities. Deciding what should be prioritized remains the responsibility of people and organizations.

    For this reason, AI-generated suggestions and team decisions should be distinguished in meeting records. Teams need to document which questions were entered, what AI proposed, which parts the team accepted or rejected, and why. This does not mean that every conversation must be preserved in exhaustive detail. It means that the key assumptions, verification results, and final decision-maker that influenced the decision should be traceable.

    Teams should also establish a standard procedure for reviewing AI-generated responses. They must determine whether the sources of facts and figures have been verified, whether the information is current, whether it matches the organization¡¯s actual conditions, whether any stakeholders have been omitted, and whether the response contains biases that disadvantage particular groups. The more important the decision, the more thoroughly it should be cross-checked against original sources, expert opinions, and information from customers and frontline operations. The level of verification should never be reduced simply because AI expresses itself confidently.

    It is also important to develop the habit of indicating uncertainty. Distinguishing confirmed facts from estimates, assumptions, and opinions reduces the risk of placing excessive trust in AI responses. Market size, for example, may be confirmed through official statistics, whereas a competitor¡¯s future response is closer to an estimate. Customer reactions may also be partially inferred from historical data, but they cannot be determined with certainty without actual research. Clearly stating these distinctions allows teams to identify which areas require additional verification.

    Rules for team AI use should vary according to the level of risk involved in the work. Relatively simple review may be sufficient for brainstorming ideas or organizing meeting titles. Far stricter standards are required for matters involving personnel decisions, investments, legal issues, safety, healthcare, or personal information. Organizations must specifically define the scope of data that may be used, whether human experts must review the output, who has approval authority, and how records will be retained.

    Teams should also accumulate successful prompts and failed cases as shared assets. They need to record not only which questions produced useful answers, but also which questions biased the discussion or omitted important conditions. Even a brief post-meeting review of whether AI broadened participation, fragmented the discussion, or improved judgment can help a team develop its own principles of use. The ability to use AI is not a skill completed through a single training session. It is a collaborative capability developed through repeated experimentation and reflection.

    The leader¡¯s role also changes. A leader does not need to be the person who uses AI most skillfully on the team. More importantly, the leader must create an environment in which members can freely ask questions and raise objections, while preventing AI responses from becoming unquestioned authority. The leader should monitor whether speaking power is becoming concentrated among technologically proficient members, whether field experience and minority opinions are being excluded from the input process, and who is accountable for the final decision.

    Using AI at the team level is not a project for automating meetings. It is an effort to redesign the structure of collaboration. When individuals use AI alone, speed is the most visible benefit. When teams use it together, relationships, authority, participation, and responsibility determine performance. Even if AI produces a perfect meeting summary, it cannot be considered good collaboration if team members were excluded from the discussion. Even if meeting time is reduced, it is difficult to claim that productivity has improved if the team failed to examine flawed assumptions.

    Strong teams do not treat AI as an entity that thinks on their behalf. They use it as a mechanism that helps members articulate their ideas more precisely, compare different perspectives, and question familiar judgments. When AI presents an answer, the team asks about its evidence, identifies missing conditions, and examines opposing possibilities. Once a decision has been made, members should not say, ¡°AI chose it.¡± They must be able to explain why the team made that choice.

    Transforming AI from a personal tool into a team tool represents a much greater change than simply having multiple people use the same technology. It means building a new way of collaborating in which questions are formulated collectively, AI is assigned diverse roles, responses are verified together, and responsibility for judgment remains with humans until the very end. When these principles become established, AI can serve not as a technology that diminishes the voices of team members, but as one that connects a wider range of perspectives. It becomes not a tool that replaces the team¡¯s thinking, but a shared intellectual resource that helps the team think more deeply and broadly.

    Reference
    Harvard Business Review, May 2026, Rosani, G., Farri, E., Trabucchi, D., and Buganza, T., It¡¯s Hard to Use AI as a Team. These 3 Practices Can Help