You can't remember how to freeze the top row of a spreadsheet. Ten years ago you would have turned to the person at the next desk: "Hey, how do you do this?" They would have shown you, and while they were at it, mentioned that their department is switching software and that the team down the hall is hiring. Five years ago you would have typed the same question into the team chat. Today you will probably ask a chatbot and have the answer in four seconds.
The answer is the same. What is gone is the "while they were at it". Sociologists have had a name for that part for a long time, and it helps to see what exactly is changing.
Acquaintances, not friends
In 1973 the American sociologist Mark Granovetter published a paper called "The Strength of Weak Ties". Strong ties are close friends and family. Weak ties are acquaintances: a colleague from another floor, a former classmate, someone you once chatted with at a conference.
Granovetter interviewed people in a Boston suburb who had recently changed jobs through a personal contact. There were 54 of them, all professional, technical or managerial workers. Only 16.7% saw the person who told them about the job often, meaning at least twice a week. The rest saw that person occasionally, or once a year or less.
The explanation is simple. Your close friends know roughly what you know, because you move in the same circle. An acquaintance lives in a different circle and brings news that yours has not heard yet.
For half a century this was an influential idea that nobody had tested in an experiment. In 2022 a study in Science did. Researchers used LinkedIn's experiments with its contact-recommendation algorithm, covering more than 20 million people over five years. Weak ties did lead to new jobs more often than strong ones. With a refinement: moderately weak ties worked best, not the very weakest, and the effect was stronger in digital industries.
The first wave: an office without a corridor
Few people build weak ties at work on purpose. They happen at the coffee machine, in the lift, in a meeting you joined by mistake. So their first serious test came in 2020, when offices emptied.
The most detailed measurement comes from Microsoft. Researchers analysed the emails, calendars, messages and calls of 61,182 of the company's US employees over the first half of 2020. Moving the whole firm to home working made the collaboration network more static and more closed within its own groups, with fewer bridges between departments.
By the authors' estimate, people spent about a quarter less of their time collaborating with colleagues from other groups. Real-time conversation went down, written messages went up, and new contacts were added more slowly.
The limits matter. This is one company, in the first months of a pandemic, and nearly all the authors worked there. The study does not show what happened to those networks afterwards.
Working from home has not disappeared, though it never became the norm. According to Eurostat, for a decade before the pandemic about 5% of employed people in the EU usually worked from home. In 2025 the figure was about 9%, and it has held there since 2022. Countries differ enormously: roughly one in five in Finland, just over 1% in Romania.
The second wave: an answer without a colleague
Remote work removed the corridor. Generative AI seems to touch something else: the reason to approach a person at all.
MyIQ, a cognitive-assessment platform, surveyed more than 22,000 employees across ten regions in June and July 2026. Among those who use AI regularly, 74% now put questions to a chatbot that they would once have asked a colleague. 59% ask colleagues for a second opinion less often, and 48% say spontaneous conversations during the day have become rarer.
Clutch, a business-services marketplace, surveyed 669 US workers who use AI regularly in October 2026. Nearly half, 45%, take a work question to AI first. 16% go to a coworker first, 9% to a manager.
A third source is the maker of one of these tools. In August 2025 Anthropic surveyed 132 of its own engineers and researchers and ran 53 interviews. Its report says the chatbot has become the first stop for questions that used to go to colleagues, and that some junior staff bring fewer questions to senior ones.
Where the data contradict themselves
Stop here and the story is simple: AI pulls people apart. But the same surveys also show the opposite.
In the Clutch survey, 57% say AI has created silos on their teams. In the very same survey, 90% say they collaborate with coworkers as much as or more than before AI, and 91% say their relationships with colleagues are unchanged or better. It is not obvious how both can be true of the same office.
In Anthropic's report, about half of the people interviewed noticed no change at all in how their team works together.
And above all, these are self-reports. People are describing how they think their behaviour has changed. Data on actual behaviour are scarce. The closest study we found looked at open-source developers on GitHub: after getting access to the Copilot coding assistant, they shifted towards working on their own rather than with others. But it is a 2024 working paper, two of its authors worked at Microsoft and GitHub, and an open-source project is not an office. We found no study that has counted, inside a company, the emails and conversations before and after AI arrived, as was done at Microsoft for remote work. The MyIQ survey itself states that its findings do not establish that AI use directly weakens workplace relationships.
It is also worth remembering who is asking. MyIQ sells cognitive tests and training, Clutch connects businesses with service providers, and Anthropic is studying its own product. That does not make the numbers wrong, but independent academic measurements are still missing.
What appears in its place
The small question to a colleague always had a price, and not everyone found it equally easy to pay. In the Clutch survey, 76% had at least once asked AI a work question they were too embarrassed to ask a person — most often how to do a task or what a term means.
For a newcomer, for someone working in a second language, for anyone afraid of looking incompetent, being able to ask without witnesses is a real change. Some people in Anthropic's report say they value no longer taking up colleagues' time. In the MyIQ survey, 71% say they have become more self-sufficient at work.
The same people can also see what may be getting lost. In the MyIQ survey, 46% say they know less about how their colleagues think and approach problems. In the Clutch survey, 76% worry that newer employees are missing out on learning from colleagues, and one in three has noticed less mentoring.
And respondents draw a line themselves. When it comes to knowledge about this particular company — who is responsible for what, why things are the way they are, whom to go to — 67% in the Clutch survey trust a coworker over AI.
What we don't know yet
Granovetter's theory tells us where to look for consequences, not whether they will come. If small questions really were an unnoticed channel through which news and opportunities moved between departments, narrowing it will show up late, and not in surveys: in how people find jobs and ideas several years from now.
The other possibility is that "how do I freeze a row" was never that bridge. Perhaps weak ties rest on shared projects, lunches and chance encounters, and the chatbot takes over only the part of communication nobody will miss.
Finally, almost all of the AI data come from the United States or from office workers who already use AI heavily. What this looks like in other countries and other kinds of work is still unknown.
This is the first article in the “AI and work” series. The next one is coming soon.
Questions for discussion
- Think of the last work question you asked a person rather than a search engine or AI. What did you get besides the answer?
- Are there questions you deliberately take to a colleague even though a chatbot would answer faster? Why?
- If a newcomer can ask AI about everything, what should a team do so that they still get to know people — and should it do anything at all?
- Where did the weak ties in your own life come from, the ones that later turned out to matter?
Log in or Sign up to join the discussion.