AI Collaboration: Relationslipping and Team Guidelines
Relationslipping is the weakening of work relationships as interactions with colleagues are replaced by AI. Related survey figures can be understood in the context of the environment for asking questions, asynchronous communication, and standards for sharing how AI is used.
Relationslipping is a phenomenon in which work relationships within a team weaken as people repeatedly turn to AI instead of colleagues.
In a survey featured by HBR, work coordination was 22.3% lower in teams where replacing colleagues with AI was common.
49% of respondents said they experienced colleagues turning to AI instead of them at least sometimes.
Psychological safety and consistent team communication help create an environment where people can ask colleagues questions.
When evaluating the results of AI use, contributions to knowledge sharing need to be considered alongside individual speed.
Even if AI helps you finish work faster, your ability to collaborate with colleagues can weaken. Relationslipping is the weakening of work relationships as people turn to AI instead of colleagues. Team management should address both the environment for asking questions and the ways knowledge is shared.
The survey figures in this article are based on an HBR research overview published on October 15, 2026.
What is Relationslipping?
Relationslipping is the weakening of work relationships as AI replaces interactions with colleagues. The English term used by the researchers is Relationslipping. The key is not how much AI people use, but how their conversations with colleagues change.
When you ask a colleague a question, you learn about their experience as well as getting an answer. As you listen to their explanation, you learn the criteria they use to assess a problem. These experiences help you decide whom to ask for help next time.
HBR's research overview explains how relationships weaken. Skipping conversations reduces opportunities to discover a colleague's expertise. When AI takes over the work process, it also becomes harder to understand a colleague's judgment. This explanation is available in the summary of HBR's research overview.
What the survey figures show
The survey shows an association between replacing colleagues with AI and team relationships. It covered about 1,000 full-time office workers on project teams in the United States. The findings were presented in HBR in October 2026.
Survey measure | Reported figure | What to distinguish when reading
Work coordination in teams where replacing colleagues with AI is common | 22.3% lower | A finding about work coordination, not a rate of decline in output
Team trust under the same conditions | 21.9% lower | A finding about trust among colleagues
Respondents who had experienced colleagues turning to AI instead of them at least sometimes | 49% | Not the percentage of people who were laid off or whose roles disappeared
Likelihood of replacing colleagues with AI among members with low psychological safety | At least twice as likely | An association with the behavior of turning to AI instead of colleagues
Work coordination among AI users who stay in regular contact with their team | 28% higher | No guarantee that contact alone will produce the same effect
Team trust among AI users who stay in regular contact with their team | 19% higher | An observed association between communication and trust
These figures also appear in BetterUp's post introducing the research. The public summary alone does not make it easy to verify all the comparison groups and formulas. The survey dates must also be distinguished from the publication date.
Differences found in a survey do not, by themselves, establish causation. The possibility that teams with weaker relationships turned to AI more often must also be considered. Do not use these figures as effect sizes that apply to every organization.
Comparing individual productivity and team collaboration
Individual productivity focuses on the ability to finish one's own work. Team collaboration includes the process of connecting people's knowledge. It is therefore difficult to judge team performance solely by the speed of individual work.
Category | What to check in individual work | What to also check in team work
Getting answers | Whether the person quickly obtained the answer they needed | Whether they know colleagues with relevant experience
Producing work | Whether the person completed a draft or analysis | Whether colleagues can understand the reasoning behind it
Solving problems | Whether the person resolved what was blocking them | Whether someone else will have to solve the same problem again
Experience using AI | Whether the person found an approach that works for them | Whether other members can also use that approach
This table applies the concerns raised by the research to team reviews. It is not a validated assessment scale or diagnostic tool. Choose items that fit your team's work and use them as a basis for discussion.
Psychological safety and barriers to asking questions
Psychological safety is the sense that you can ask questions or reveal mistakes without worrying about negative consequences. It is hard to approach colleagues if you worry that asking questions could hurt your evaluation. AI can provide someone to ask without this interpersonal burden.
If you are a leader, start by examining how you react when someone asks a question. A sigh or a dismissive attitude can make people hesitate to ask the next question. If you cannot answer right away, one option is to let them know when you can respond.
· Identify where the person is stuck before judging the quality of the question.
· When you do not know the answer, point them toward someone or something you can check together.
· When you disagree, explain the reasoning behind your judgment.
· Do not assume that asking for help is a sign of incompetence.
What to include in asynchronous communication
Asynchronous communication is a conversation that can continue without everyone being online at the same time. Reading and replying later to messages left in Slack or email are examples. The researchers suggest staying in touch with colleagues through these channels.
Posting only whether work is complete may not reveal where judgment is needed. Encourage people to share questions they are still working through. It helps to specify where a colleague's experience is needed.
· Where you are currently stuck and what you have already checked
· Parts of AI responses you have not yet been able to assess
· New facts learned during the work and the evidence behind them
· Questions that need a colleague's experience or review
Do not turn a requirement for daily contact into an obligation to be constantly available. Asynchronous channels are a way to communicate without having to respond immediately. Message counts alone also make it difficult to assess the level of collaboration.
Guidelines for sharing the process of using AI
When sharing AI output, include a record of the judgments made by people. A completed analysis alone makes it difficult to know which questions were effective. Document the process so colleagues can refer to it when doing the same work.
Using AI for customer analysis is one example. The items below are suggestions for preserving the results as team knowledge. They are not examples of actual business performance or verified time savings.
Item to share | What to record | What colleagues can learn
The problem you were trying to solve | The purpose of the analysis and the questions that needed answers | The work context in which to apply the method
Requests made to AI | Questions and conditions that influenced the result | Ways of making requests that they can use
Changes made by people | Responses that were accepted or excluded | The process of reviewing AI responses
Reasons for decisions | Facts or work criteria used to make changes | Why the result was accepted
Remaining questions | What has not yet been verified | What needs further review
Saving only prompts may leave out why the results were accepted. Recording the reasons for changes also gives colleagues clues to understand the judgment behind them. Check whether other people have used these records as well.
Criteria for evaluating the results of AI use
Evaluations can include both individual work results and contributions left for the team. Checking only whether someone uses AI skillfully makes it difficult to assess knowledge sharing. Also look at whether they have documented their experience in a form colleagues can use.
The following are examples of applying the researchers' knowledge-sharing suggestions to evaluations. They are not criteria with specific scores or weights. Choose what to observe based on how your team works.
· Whether the person documented work methods that other members can use
· Whether they explained to colleagues why they changed AI responses
· Whether they contributed to colleagues' questions or review requests
· Whether the shared methods were used again in actual work
Common mistakes and misreadings of the figures
The most common misunderstanding is to read survey figures about relationships as measures of work performance. The same number means something different when the survey measure changes. Keeping the following distinctions in mind can help avoid overstating the findings.
Misunderstanding | Accurate distinction
If collaboration is 22.3% lower, output falls by the same amount | Work coordination and output are different measures
49% lost their jobs because of AI | The percentage who experienced colleagues turning to AI instead of them
Sending messages every day increases trust by 19% | An association between communication and trust, not a guaranteed improvement
People who use AI a lot cause relationships to weaken | The issue concerns how AI use replaces interactions with colleagues
Posting only the output also shares the know-how behind its use | How questions were asked and why changes were made must be documented separately
Where to start with a team review
Start by identifying conversations that have decreased in recent, actual work. It is easier to discuss a single task than a vague impression of the whole team. The sequence below is an operational suggestion based on the concerns raised by the research.
· Choose one recent task that involved AI. Identify the problem you were trying to solve.
· Check whether any conversations with colleagues were skipped in the process. Review whether you missed experience you needed.
· Record the human judgment applied to the AI output. Leave it in a channel colleagues can refer to.
· Check whether the record was used in the next similar task. Adjust how you share any parts that were not helpful.
Injoys
