There is a new condition spreading through hybrid teams, and it does not look like a problem. It looks like a win. Output metrics are up. Pull requests are shipping. Decks are landing. Engineers, PMs, designers, and ops leads have all figured out how to do more in less time. Dashboards are green. Retention is fine.
Then a real fire starts. A launch slips. A customer escalates. A senior person leaves and three projects lose continuity overnight. The team that looked strong on paper cannot coordinate its way out of the meeting room. Shared context is gone. The weak ties that used to carry information across the org have frayed, and nobody can name when it happened.
This is what I call hyper-productive silos: hybrid teams where each individual has become more productive, mostly with AI, while the team as a unit has become less capable. The brilliant engineer who used to mentor two juniors over Slack now opens Copilot first and never types the question. The PM who used to ping engineering with a half-formed idea now runs it past ChatGPT and arrives at standup with a finished answer nobody helped shape. Each person is faster. The team is slower.
A 6-Question Diagnostic for Managers
Most managers cannot see this pattern until something breaks. The individual metrics keep glowing. Headcount is steady. Slack volume might even be up. The signals you need are behavioral, and they live in the negative space: things that used to happen and have stopped. Here is the diagnostic I give managers who suspect their hybrid team is silently siloing.
- Response-time decay on ambiguous questions. Crisp, well-scoped questions still get answered fast. Vague, half-formed, "is this stupid?" questions sit longer than they did six months ago, or never get asked in public at all.
- Async absences during shared time. During core hours, more people are heads-down in 1:1 work with an AI assistant. Group channels are quieter. The hallway chatter equivalent has thinned out.
- The mentorship drop. Senior engineers report fewer informal questions from juniors. Juniors report that they "just ask the model first." Treat that as a leading indicator.
- Knowledge-base growth without team-capability growth. Your Notion, Confluence, or Coda is exploding. The team's ability to handle a novel situation as a unit has not budged.
- Solo decisions where there used to be a thread. Decisions that used to involve two or three people pinging each other now show up fully-formed in a doc, often with an AI co-author no one acknowledges.
- Missing escalation moments. A few months ago, someone on this team would have flagged a small problem early. Now the same problem surfaces as a crisis. The weak-tie early-warning system has gone quiet.
Three or more of these in the same team is the pattern. Five or six is the pathology.
Why Anchor Days Hide the Problem
The hybrid response to "is the team okay?" has been anchor days: two or three days in the office, the rest distributed. The research from Nick Bloom and collaborators is broadly positive on retention and short-run output (Bloom et al., NBER 2024). Anchor days do real work for hybrid teams when used well.
They also create an illusion. On anchor days, in-person hours look healthy. Lunches happen. Standups are crisp. The leader walks the floor and sees a team. The other three or four days, when silo behavior actually runs, are invisible to anyone above the IC level.
Microsoft's 2025 Work Trend Index on the "infinite workday" sharpens the off-anchor picture. Knowledge workers are interrupted on average every two minutes, with around 275 disruptions per day across chat, mail, and meetings. Forty-eight percent of employees and 52% of leaders say work feels chaotic and fragmented (Microsoft 2025 WTI). Fifty-seven percent of meetings are ad-hoc, with no invite.
This is a chemistry problem disguised as a focus problem. Fragmentation makes individuals reach for the fastest possible answer, which is increasingly an AI assistant, which routes the chemistry-building micro-interactions (the questions, the heads-up, the quick poke) out of the team entirely. The anchor day comes around and the metrics look normal. The other three days, the team is dissolving.
The Shame Loop That Keeps the Pattern Hidden
None of this would matter if people talked openly about how they were using AI. They are not.
The Microsoft and LinkedIn 2024 Work Trend Index found that 75% of global knowledge workers are now using generative AI at work, doubling in six months. Seventy-eight percent are bringing their own AI tools to the job. The piece that breaks the team is downstream of that: 52% of users are reluctant to admit using AI for important tasks, and 53% worry that being seen as an AI user makes them look replaceable (Microsoft + LinkedIn WTI 2024).
Slack's Workforce Index Fall 2024 (n=17,372) made the social cost more explicit: 48% of employees are uncomfortable telling their own manager they use AI for core work. The reasons employees gave were the heart of the problem: being seen as cheating, lazy, or less competent (Slack Workforce Index, Fall 2024).
So the modal hybrid worker is now: using AI for most of their work, hiding it from their manager, hiding it from teammates, and outsourcing exactly the kinds of small, exploratory, "let me think out loud" questions that used to build team chemistry. The output looks great. The shame loop has eaten the connective tissue.
The Klarna Case Study
Klarna is the clearest public arc of what hyper-productive silos look like at company scale. In February 2024, the company announced a high-profile OpenAI partnership and effectively replaced 700 customer-service roles with an AI agent. Coverage was glowing. Headcount went down. Cost-per-ticket went down. Volume handled went up.
By May 2025, CEO Sebastian Siemiatkowski publicly admitted the company "went too far." Quality had dropped. Customer experience had degraded in ways the dashboards had missed. Klarna began rehiring humans into customer service and re-positioned as "hybrid AI plus human" (Fortune, May 2025).
The Klarna arc is the macro version of the team-level pattern. Each individual interaction was more productive. The aggregate experience was worse, because the parts of the work that depended on judgment, escalation, and read-the-room were the parts the AI handled worst. Nobody noticed until customers did. Klarna's own framing is more diplomatic and the press treatment is heavily filtered, but the direction is undisputed: a company optimized hard on per-task productivity and discovered the team-level capability had hollowed out underneath it.
What the Research Says About Weak Ties
The research on why this happens is older than the AI moment, and stronger than most leaders realize. In 1973, the sociologist Mark Granovetter published "The Strength of Weak Ties" and showed that the casual, peripheral connections in a network are what move new information across an organization (Granovetter 1973). Weak ties are how teams stay aware of what other teams are doing, how juniors find unwritten norms, how problems get flagged before they become crises.
The first big causal evidence that remote work erodes weak ties came in September 2022, when Longqi Yang and collaborators at Microsoft published a study of 61,182 Microsoft employees in Nature Human Behaviour. After the firm went fully remote, cross-group collaboration decreased, weak ties decreased, and communication shifted from synchronous channels to async media (Yang et al., 2022). The company building Copilot watched its own cross-group connective tissue thin in real time.
Layer the AI evidence on top. Brynjolfsson, Li, and Raymond's NBER study of 5,179 customer-support agents found that generative AI raised average productivity by 14%, with the biggest gains (+34%) for novices and essentially zero gain for experienced workers (NBER w31161). The implication: juniors look more senior on routine tasks, without doing any of the senior-judgment reps. The mentorship feedback loop that used to create senior judgment is exactly what the team has stopped doing.
The Dell'Acqua BCG study of 758 elite consultants is the other half. Inside the "jagged frontier" of what AI is good at, consultants got 12.2% more tasks done, 25.1% faster, with 40% higher quality. Outside that frontier, working alone with AI, they produced outputs 19% worse than peers without AI (Dell'Acqua et al., HBS 24-013). Even elite individuals make worse decisions when isolated with a confident model.
And Lee et al. at CHI 2025 (Microsoft and CMU, 319 knowledge workers across 936 reported tasks) found that higher confidence in the AI's output correlated with less critical thinking, with workers describing their own judgment as "atrophied and unprepared" on harder edge cases (Lee et al., CHI 2025).
No peer-reviewed study has yet tested "AI directly causes team-chemistry decay" as a single hypothesis. The argument here synthesizes three literatures: remote-work network effects, AI productivity effects, and judgment-atrophy effects. The convergence is hard to ignore.
The connective tissue of a team is built from thousands of low-stakes interactions. Hyper-productive silos remove the interactions and keep the output. Information silos in the classic sense were about teams that did not share data. The new hybrid version is about individuals who do not need to share because the model already answered them.
The Counter-Argument, Taken Seriously
Microsoft's 2025 Work Trend Index argues that "frontier firms" (companies pairing humans with AI agents in tightly designed workflows) outperform peers. The data shows real gains where the design is intentional. The same report describes the infinite workday and the fragmentation. Both are true. Frontier-firm gains are individual amplification effects. Infinite-workday losses are team-coordination effects. Treating them as one story is the trap most leaders fall into.
Bloom's hybrid research is similarly real. Anchor days, well-implemented, are a net positive on retention and short-run output. But Bloom measures retention and output. Team chemistry under load is a separate variable, and anchor days were never designed to protect what breaks during a crisis.
The Brynjolfsson "levels the playing field" finding is the most-cited piece of optimistic AI evidence. Juniors do catch up on routine work. The same study shows experienced workers gain almost nothing, which means the senior judgment juniors are leapfrogging is still the part that matters when things get hard.
The macro number from McKinsey 2025 should keep leaders honest: roughly 90% of companies have deployed AI, and 94% report no significant value from it (McKinsey 2025). The simplest explanation is that the individual gains are being eaten somewhere, and the most likely place is the team layer no dashboard measures. Gallup's 2025 State of the Global Workplace puts a price on it: global engagement fell from 23% to 21% in 2024, the second drop in 12 years, driven mostly by managers, with $438B in lost productivity globally (Gallup 2025). Atlassian's 2024 State of Teams report estimates 25 billion hours lost annually to ineffective collaboration, with 64% of teams lacking shared goals (Atlassian 2024).
These are the same pathology measured at different altitudes. AI brain fry is the individual symptom. Hyper-productive silos are the team symptom. Teaming is what AI cannot do, and it is exactly the muscle most hybrid orgs are now under-training.
The Fix Is Structured, Repeated Team Practice
Anchor days do not rebuild weak ties by themselves. Slack volume does not. AI assistants will not. What builds team chemistry has always been the same thing: repeated, low-stakes, shared, slightly-hard moments where the team has to think together as a unit. Most hybrid teams do not get those anymore because nobody designs them in. The default workday has Copilot, ChatGPT, async docs, ad-hoc meetings, and a manager too busy to notice the connective tissue is thinning.
QuestWorks is built to make that slot non-negotiable. Twenty-five minutes a week, two to five players per quest, dynamic grouping so people are not always working with the same teammates, AI-facilitated so the manager does not have to run it. Quests run on QuestWorks' own cinematic, voice-controlled platform, so the work happens away from the AI tools causing the silo problem. Slack or Teams handles install, invites, onboarding, leaderboards, and HeroGPT coaching, and those HeroGPT conversations stay private to the player.
The point of running it weekly is the accumulation. Repeated shared experience builds the weak ties Granovetter described. Repeated behavioral practice creates the senior judgment Brynjolfsson's juniors are skipping. Repeated 25-minute moments where the team has to talk, decide, and act as a unit rebuild the chemistry the infinite workday is corroding. The goal is to give your team a reason to be a team again, on a schedule, in a format managers do not have to manufacture. That is the lever that prevents team intelligence from slipping while individual productivity climbs.
The teams that win the next two years will be the ones that kept the weak ties alive while everyone else was busy looking productive.
