A Team in Name Only
The textbook definition of team comes from J. Richard Hackman: a bounded set of people, working interdependently toward a shared goal, with stable membership and collective accountability. Hackman validated this through the Team Diagnostic Survey across 2,474 members and 321 teams, and his 64-team study of US intelligence community groups found that six conditions, beginning with "real team" (bounded, stable, interdependent), accounted for 74% of variance in effectiveness (6 Team Conditions). That definition held for half a century.
It holds less well in 2026. Look at the actual state of a typical knowledge-work group on a Tuesday. Microsoft's 2025 "Infinite Workday" research, drawing on telemetry across millions of Microsoft 365 users, found workers are interrupted every 2 minutes during core hours by a message, mention, or meeting. They receive 117 emails and 153 Teams messages a day. Fifty-seven percent of meetings are ad-hoc, scheduled with less than an hour's notice. Meetings after 8 PM are up 16% year over year. Forty-eight percent of employees and 52% of leaders describe their workday as chaotic and fragmented, and 80% say they lack the time and energy to do their jobs (Microsoft WorkLab).
Look at the roster. GitHub Copilot's autonomous coding agent completes 75% of pull request jobs in under 18.5 minutes, where the human counterparts take more than half a day, and the agent submits work that matches the human distribution of feature development and bug fixes (arXiv 2507.15003). Atlassian's Rovo, pitched as an "AI teammate," reports that 87% of Rovo Chat users say it moves work forward faster (Atlassian). Microsoft's 2025 Ignite announcement framed it bluntly: every team, project, and meeting eventually has an AI teammate (Microsoft). Salesforce's Agentforce sits in the same category.
And look at the network. Yang et al.'s 2022 Nature paper, drawing on 61,182 Microsoft employees through the shift to remote work, found a 32% reduction in time spent with weak ties and showed networks becoming more static and siloed (Nature Human Behaviour). The serendipitous overlap that produced team boundaries in the office is structurally weaker now.
By Hackman's definition, many groups carrying the label "team" in 2026 are teams in name only. Boundedness is fuzzy. Stability is gone. Interdependence is broken by interruption. Shared accountability is muddled by AI co-authors. The container is leaking. The question is what holds.
Three Forces Stressing the Definition
The classic definition is being stressed along three specific dimensions, and each maps to research that is fresh in 2025 and 2026.
Boundedness is eroding. Amy Edmondson named this trend in her 2012 Harvard Business Review article "Teamwork on the Fly," introducing the verb "teaming." Her canonical case was the 2010 Chilean miner rescue: 33 trapped miners, NASA, Chilean Navy, US drillers, no prior history together, succeeding under extreme pressure. Edmondson reframed teamwork as a dynamic activity rather than a bounded entity. In 2026, agentic AI accelerates the same trend at the individual roster level. Copilot, Rovo, and Agentforce join projects, hold context, hand off work, and depart. The Yang et al. network finding sits on top of this: when weak ties weaken, the boundary of "who is on my team this quarter" stops being obvious.
Interdependence is eroding. Hackman's "real team" requires members who depend on each other to produce work no one could alone. The infinite workday subverts this. When a knowledge worker is interrupted every 2 minutes, when 57% of meetings are unscheduled, when 117 emails and 153 Teams messages compete for attention each day, the unit of work shrinks to the gap between interruptions. Individual AI copilots make this worse on the throughput side: they let solo work get done faster, which reduces the felt need for a teammate. Microsoft's 2025 Work Trend Index found 42% of workers now choose AI over a human colleague for 24/7 availability (Microsoft WTI 2025). The economic case for interdependence weakens when a co-worker is replaced by an always-on agent for the marginal request.
Shared identity and accountability are eroding. Identity used to come free with assembly. The standup, the lunch table, the after-meeting hallway. In hybrid, those rituals are vestigial or absent. Accountability is also blurred. When an autonomous coding agent submits a pull request, who owns the consequence? Microsoft and LinkedIn's 2024 Work Trend Index found 75% of knowledge workers use AI, and 78% are BYOAI, bringing their own tools without sanction (Microsoft). Half the work happening in the team is done with tools the manager cannot see. Atlassian's State of Teams 2025 reports knowledge workers spend 25% of their time searching for answers, an estimated 2.4 billion hours wasted per year across the Fortune 500, and 71% admit they are not maximizing the AI tools available to them (Atlassian). The team narrative, the "what we did together this week," is harder to assemble.
What Endures
The empirical core of team science is not breaking. It is being stressed.
Hackman's six conditions, real team, compelling direction, enabling structure, supportive context, expert coaching, sufficient resources, still explain 74% of variance in his intelligence-community study and 60% to 74% in subsequent validations. The Team Diagnostic Survey still works.
Psychological safety still tops the list. Google's 2015 Project Aristotle, across 180 teams, found psychological safety the strongest predictor of high performance (re:Work). Edmondson and Mortensen's 2021 HBR follow-up argued the construct must extend beyond work content in hybrid settings, into work-life integration and the texture of asynchronous communication (HBR).
Anita Woolley's 2010 Science paper on the collective intelligence factor (the c-factor) still holds. Team-level cognitive performance is predicted by social sensitivity and equal turn-taking, not by the IQ of the smartest member (Science). Sandy Pentland's sociometric badge research showed communication patterns alone, without semantic content, predict team performance. The signature of a high-performing team is structural, not lexical.
Mark de Rond's 2012 book There Is an I in Team, drawing on ethnographic work with elite rowing and surgical teams, rebuts the cozy Tuckman story. Tuckman's 1965 forming-storming-norming-performing model is popular but was never empirically re-validated as predictive. De Rond's finding is the opposite of Tuckman's implication: harmony is a consequence of performance, not its cause. Productive tension is often the engine (Cambridge JBS).
None of this collapsed when Slack arrived. None of it collapsed when COVID forced remote. None of it is collapsing now under agentic AI. The conditions still hold. They are just no longer free byproducts of co-location.
What Needs Rebuilding
The thing that has degraded is the ritual scaffolding. Co-located teams got identity, witness, repetition, and shared narrative for free. The office supplied the recurring rhythm. The whiteboard supplied the witness. The hallway supplied the unstructured time that produces weak-tie reinforcement. Hybrid distance and AI interruption strip those out.
What remains is engineered substitutes. Slack's June 2025 Workforce Index, surveying tens of thousands of desk workers, found daily AI users report 64% higher productivity, 58% better focus, 81% greater satisfaction, and (most relevant here) are 246% more likely to feel connected to colleagues (Slack). That 246% is not because AI is friendly. It is because the daily AI user is, in practice, embedded in workflows that surface what teammates are doing, who is contributing, and what got finished. Connection has been engineered into the workflow, replacing the ambient version the office used to provide.
The four ingredients that need rebuilding are concrete:
Repetition. A recurring rhythm with the same small group. Not the all-hands. Not the cross-functional review. A bounded cohort meeting on a cadence small enough that participants notice if someone misses, and consistent enough that it becomes part of how the week is shaped.
Witness. Members seeing each other do work in real time. The watercooler's actual function was incidental witness: seeing a colleague handle a hard customer, defuse a tense moment, ship a fix. Asynchronous standups in Slack do not produce witness. Live coordination under pressure does.
Shared narrative. A story about who the team is, what it survived together, what it is becoming. Co-location supplied this through inside jokes, shared crises, and the cumulative memory of having been in a room together. Without engineered narrative-building, hybrid teams are left with only an org chart.
Role legibility. Who does what, who is strong at what, who to ask when. The office made role legibility ambient: you watched and learned. In hybrid, role legibility has to be made explicit. Public profiles, visible strengths, clear escalation paths.
The Redefinition
A team in the hybrid AI era is best defined as a deliberately rehearsed group of humans (and increasingly agents) maintaining shared identity, interdependence, and accountability through engineered repetition rather than co-located default.
The deliberate rehearsal is the load-bearing phrase. In the old definition, teamness was the byproduct of assembly. In the new definition, teamness is a property that must be actively manufactured against the entropy of distance, interruption, and roster fluidity. The team does not exist because people share a manager or a project. It exists because the group keeps showing up to engineered repetitions of coordination that produce identity, witness, and narrative as outputs.
This also resolves the agentic AI question without contorting Hackman's framework. The roster includes agents because agents take work, hold context, and influence outcomes. The agents are not bounded in the human sense, but they participate in the rehearsal. The accountability question is answered by a different mechanism: humans retain accountability for what the agents produce, and the team is the unit that owns the consequence. This is consistent with Edmondson's teaming-as-verb: coordination across whatever entities are present, repeatedly, under pressure.
Counter-Arguments
Three objections deserve direct engagement.
"Teams are over. Networks are the unit now." The network-replaces-team argument shows up in every wave of organizational change. The Yang et al. data complicate it. Distributed work pushed people deeper into their strong-tie clusters, cut weak-tie time by 32%, and produced more siloed networks rather than richer ones. The pure-network model loses the very serendipity it claims to enable. The team unit is not dissolving into a network. It is being re-engineered because the network, left alone, contracts.
"Redefining team around AI is premature." Daron Acemoglu's NBER w32487 estimates total factor productivity gains from AI of less than 0.66% over a decade, possibly less than 0.53%, undercutting the louder vendor claims (NBER). Brynjolfsson, Li, and Raymond found 14% average productivity gains from generative AI, with 34% for novices, in a real-world deployment (NBER 31161). The honest read is that the macro curve is flatter than vendors claim, but the micro shift in how individual work gets done is real and is changing what a team's day looks like. Waiting for macroeconomic clarity before redefining the team is a luxury most managers do not have.
"Hackman's conditions still hold; nothing has changed." The conditions do still hold. What changed is the cost of producing them. "Real team" used to be the default of co-located work. It is now the hardest of the six conditions to manufacture. The definitional shift is not that Hackman was wrong. It is that the first condition went from free to expensive, and most organizations have not yet adjusted their team operating budget to reflect that.
What the Operational Arm Looks Like
If a team is now a deliberately rehearsed group, the practical question is what the rehearsal is. Gallup's 2025 data shows managers account for 70% of the variance in team engagement, but only 44% of managers globally have any formal training, and engagement fell to 21% in 2024 (Gallup). The variance lever exists, but the rehearsal infrastructure is missing.
QuestWorks is the operational arm of the redefinition. The category is Team Intelligence, powered by play. It runs on its own cinematic, voice-controlled platform and works with Slack as the integration layer for install, invites, and the in-Slack HeroGPT coach. A 25-minute weekly quest, 2 to 5 players, AI-facilitated, gives the bounded cohort its engineered repetition. The 9 HeroTypes are public profiles, supplying the role legibility hybrid work no longer produces ambiently. QuestDash shows strengths-based XP highlights, the visible record of who did what this week, supplying witness and shared narrative. A separate Weekly Team Health Report goes to leaders. HeroGPT coaching is private and never shares upstream. Participation is voluntary and is not tied to performance reviews.
Pricing is a flat $99 per team per month; early customers lock in $99 forever, with the standard price rising to $149, and a 10-day free trial. The economic logic is straightforward: if the team is now a deliberately rehearsed unit rather than a co-located default, the operating cost of that rehearsal has to land somewhere. The choice is whether it lands on a quarterly offsite, an unscheduled meeting at 9 PM, or a 25-minute weekly engineered rehearsal that runs on autopilot. Related reading: Why Teaming Is the Skill Gen AI Can't Replace, Team Intelligence: A New Category, What Is Team Intelligence?, and AI Brain Fry in Engineering Teams.
Hackman's six conditions still predict 74% of the variance. The work in 2026 is producing the first condition (real team) when the world no longer hands it to you. Engineered repetition is what that looks like.
