The senior engineer who stopped mentoring
Picture a staff engineer at a mid-sized SaaS company. Two years ago, when a junior on her team got stuck on a refactor, she would block off forty-five minutes, share her screen, and walk through her mental model out loud. The junior learned. She learned what the junior did not yet know. The team got a little tighter.
Today, that same engineer opens Copilot before she opens Slack. The junior pings her with a question. She glances at it, sees that GPT can probably handle the first pass, and types back: "Try asking the AI first." The work moves faster. The junior gets an answer in seconds. And one of the small, repeated acts that used to build her team has been deleted from the workday.
Multiply this pattern by every senior on every team in a hybrid economy where face-to-face nudges already disappeared in 2020. Individuals are getting faster. The team is getting hollower.
The individual productivity story is real
AI assistants, used well, deliver real individual gains.
The 2023 GitHub Copilot RCT with 95 developers found engineers given Copilot completed a JavaScript task 55.8% faster than the control group, significant at P=.0017 (Peng et al., 2023). The Harvard, BCG, and MIT "Jagged Frontier" study of 758 BCG consultants found GPT-4 users completed 12.2% more tasks, finished 25.1% faster, and produced 40% higher quality output inside the model's capability frontier. Outside it, accuracy dropped 19% (Dell'Acqua et al., 2023).
The Brynjolfsson, Li, and Raymond NBER paper tracked 5,179 customer support agents and found a 14% average lift, with novice agents up 34% and experienced agents seeing essentially zero gain (NBER w31161). Microsoft's 2024 Work Trend Index put global knowledge-worker AI adoption at 75%, doubling in six months, with 78% bringing their own AI tools whether IT sanctions it or not (Microsoft WTI 2024).
Your individual contributors are almost certainly faster than they were eighteen months ago. The question is what it is costing you.
What individual gains miss: the collaboration layer
Yang and colleagues' January 2022 Nature Human Behaviour study tracked 61,182 Microsoft employees through the firm's shift to fully remote work. Firm-wide remote work caused:
- 25% less cross-group collaboration time
- 32% less time with weak ties, the loose connections that surface novel information
- 9% fewer bridging ties overall, and 41% less collaboration time with the bridging ties that remained
- A measurable shift from synchronous conversation to async messaging
(Yang et al., Nature Human Behaviour 2022.)
Hybrid work softened the extremes and left the pattern intact. Weak ties are how teams find serendipity, surface blind spots, and discover the person two desks over solved the problem you are stuck on. When weak ties thin out, the team's collective intelligence thins with them.
Now layer AI on top of a workforce whose weak ties are already 32% lighter than pre-pandemic. The Atlassian State of Teams 2024 survey of 5,000 workers and 100 Fortune 500 executives found teams busier than ever and accomplishing less, with 64% lacking shared goals (Atlassian, 2024). The 2025 follow-up was blunter: 96% of companies reported no significant AI ROI, with siloed individual use named as the bottleneck (Atlassian, 2025).
AI as a colleague-avoidance tool
The 2025 Microsoft Work Trend Index put hard numbers on what managers had been sensing. When asked why they use AI assistants instead of asking a coworker, 8% of workers said they use AI specifically to avoid sharing credit with a colleague. Another 42% said they would feel embarrassed asking a coworker for tech help. One in four said they would not tell their organization they used AI on a task (Microsoft WTI 2025). Slack's Fall 2024 Workforce Index found 48% of desk workers are uncomfortable telling their manager they used AI, citing fears of being seen as "cheating," "less competent," or "lazy" (Slack, 2024).
Collaboration channels are shallower. AI is replacing some of the interactions that remain, and a meaningful share of workers are hiding the substitution. The team's sense of who knows what, who is struggling, and who has bandwidth gets blurrier with every silent prompt.
Klarna ran the experiment for everyone
In February 2024, Klarna announced an OpenAI partnership and a customer support AI agent that, within months, was handling 2.3 million conversations and "doing the work of 700 full-time employees." CEO Sebastian Siemiatkowski toured the press circuit.
In May 2025, Siemiatkowski admitted Klarna "went too far." Customer satisfaction had slipped, quality was lower, and the company was rebuilding a human support operation via an Uber-style flexible workforce (Fortune, May 2025). The AI cleared the bar to ship, but it could not carry the whole queue by itself. The team that used to absorb hard cases, escalate weird ones, and pass institutional knowledge between shifts had been thinned out. The damage showed up later than the savings.
Klarna is not alone. Shopify's Tobi Lütke demanded in April 2025 that managers prove AI cannot do a job before approving headcount. Duolingo cut 10% of contractors in early 2024 and writers later that year, with CEO Luis von Ahn saying headcount only grows "if a team cannot automate more of their work." IBM's Arvind Krishna told Bloomberg roughly 7,800 back-office roles could be replaced by AI within five years, and AskHR now automates 94% of routine HR transactions (Bloomberg, 2023). Each story is rational at the unit level. Each takes a bite out of the tissue that turns a roster into a team.
The mentorship pipeline is the first casualty
AI is taking over the work that used to teach junior people the business. MIT Sloan's analysis of the entry-level pipeline found that across the top 15 tech firms, entry-level hires as a share of total hiring fell by half between 2019 and 2024, from 11% to 7%. The drudgework where a junior would read every old ticket, summarize every internal doc, or draft every first version of a memo is exactly the work AI does first and best (MIT Sloan, 2025).
That drudgework is how juniors build mental models of how the company runs and where the bodies are buried. Senior-to-junior teaching is also one of the most reliable producers of cross-functional weak ties. Kill the teaching moments and you kill a meaningful share of the relationships that hold a hybrid org together. Our breakdown of AI brain fry in senior engineers covers the cognitive load side of the same coin.
The aggregate paradox: why firms cannot see the gains
If individuals are 14% to 56% more productive, where are the corporate gains?
Daron Acemoglu's May 2024 NBER paper projects that current AI deployments will add roughly 0.5% to total factor productivity over the next decade, not the 9% Goldman Sachs floated (NBER w32487). A 2025 NBER paper from Bloom, Davis, and colleagues surveyed roughly 6,000 executives across the US, UK, Germany, and Australia and found more than 80% of companies reporting no discernible impact of AI on employment or productivity at the firm level (NBER w34984).
The Microsoft and Carnegie Mellon paper at CHI 2025 added the mechanism. In a study of 319 knowledge workers across 936 generative AI use cases, researchers found higher confidence in AI was associated with less critical thinking, and that "mechanising routine tasks deprives users of routine opportunities to practice judgment, leaving them atrophied" (Lee et al., CHI 2025). Outputs were also less diverse. Across a team, that means more convergent thinking, fewer dissenting takes, less of the friction that produces good decisions. The individual is faster, the team is less varied in its reasoning, weak ties are thinner, and the firm-level gain washes out. The team layer is the missing variable in the productivity paradox.
Hybrid orgs feel this harder
The 2025 Microsoft "Infinite Workday" study, drawn from telemetry on 31,000 workers across 31 countries, surfaces the lived experience. The average knowledge worker is interrupted every two minutes, up to 275 times a day, and receives 153 Teams messages and 117 emails daily. Post-8pm meetings are up 16% year over year, with hybrid workers reporting evening hours as stress rather than recovery (Microsoft, 2025).
Gallup's 2024 State of the Global Workplace reports 20% of global employees experiencing daily loneliness, rising to 25% among fully remote and 22% among under-35 workers, and ties low engagement to an $8.9 trillion annual hit to global GDP (Gallup, 2024). AI is being asked to fix a productivity problem in a system that already has a connection problem. When the productivity tool also reduces the number of times you reach out to a teammate, the math gets worse.
What chemistry-preserving orgs do differently
The companies threading the needle share a pattern. They treat AI as a team partner rather than a personal productivity assistant, and they hold the workflow inside spaces where the team already collaborates. Atlassian's Rovo agents live inside Jira and Confluence, where the team's actual artifacts already live (Atlassian, 2025). GitLab's framing is similar: agentic patterns "live inside" the existing collaborative workflow rather than being bolted on (GitLab, 2025).
The underlying principle. If your AI deployment removes a person from the conversation, you lose a piece of team chemistry. If your AI deployment adds capability to the conversation the team is already having, chemistry compounds. Atlassian's data backs it: AI-using teams that figure out how to collaborate around the tools are 4.9x more effective than average. The 96% reporting no significant ROI are mostly running solo prompts in private windows.
Our long-read on what makes a high-performing team walks through 40 years of research that pre-dates the AI conversation. Every framework agrees team chemistry is downstream of psychological safety, shared mental models, and repeated low-stakes interaction. None of those are produced by a prompt.
The missing budget line
Most companies in 2026 have a tools budget, a benefits budget, an L&D line, and an HRIS. They do not yet have a line item for the system that builds and maintains team chemistry, because for most of corporate history that work was an emergent property of the office. People bumped into each other. Now they do not, and AI is making the bumping rarer.
This is the missing tier in the intelligence stack. Workforce intelligence platforms like Visier read aggregate workforce data. People intelligence platforms like Eightfold read individual skills. There is a third tier almost nobody serves: team intelligence, the layer where chemistry, friction, and trust are produced and measured. Our piece on why team intelligence is emerging as a category now covers the four forces driving the shift. Teaming, as a meta-skill, is becoming more valuable as individual cognitive work gets cheaper. We made that case in why teaming is the skill gen AI cannot replace.
How QuestWorks fits into the picture
QuestWorks does the team chemistry work on autopilot, so individual AI productivity gains do not come at the cost of team cohesion. It runs on its own cinematic, voice-controlled platform and works with Slack and Microsoft Teams as the integration surface. Weekly twenty-five-minute sessions place two to five players together in dynamically grouped quests where chemistry, communication, and decision-making get a low-stakes proving ground.
Underneath the play, the platform reads signal on team chemistry trends, weak-tie formation, and shared-mental-model gaps. Leaders see aggregate team health and individual strengths-based highlights. HeroGPT, the private coach, never shares upstream. Participation is voluntary and opt-in, never tied to performance reviews. Pricing is $14 per user per month for the Founder's Circle (first 50 net-new companies, locked forever) and $20 per user per month standard, with a 10-day free trial.
The bet for the next decade
Everyone is going to have the same AI. The frontier models will commoditize. The productivity ceiling for an individual augmented by a state-of-the-art assistant is converging across the industry, and the gap between a strong individual contributor and an average one with AI is narrowing fast.
The differentiator will be the team. The ability to coordinate, to trust, to disagree productively, to surface weak-tie information, to mentor across hybrid distance. Companies that treat hybrid AI deployment as a chemistry problem first will be the ones whose AI investments compound. The rest will watch the 96% no-ROI number stay fixed while their best people get faster and lonelier. The senior engineer who stopped mentoring is a leading indicator. If she is mentoring less, the system stopped rewarding it. Build a system that rewards it again, and the team gets stronger every week the AI gets better.
