The most-quoted number about AI at work this summer isn’t a benchmark score. It’s a gap. Individuals say AI is making them faster — and their organizations can’t find the gains. The tidy explanation is that the models aren’t good enough yet. There’s a better one, and it’s twenty-five years old: coordination neglect. Coined by Stanford’s Chip Heath and Nancy Staudenmayer in 2000, it names a stubborn human bias — teams are far better at dividing work up than at integrating it back together, and they chronically underinvest in the integration. Drop AI agents into a team built that way, and you don’t escape the bias. You multiply it.
That’s the argument of this piece, and it changes where you look for the fix. Lova is a chat-first AI project management product where AI agents work as first-class teammates on a shared board — each with its own identity, claiming tasks, shipping them, and advancing verifiable status alongside the humans they work with. That definition matters here because coordination neglect is precisely the problem a shared board is built to solve: it makes integration the default state of the work, not an afterthought someone has to remember. The bottleneck in 2026 isn’t how much your agents can do. It’s whether their work ever comes back together.
Key takeaways
- Coordination neglect is a documented bias: people focus on dividing labor and neglect the harder work of integrating it. It was named by Heath and Staudenmayer in 2000 and it explains the 2026 AI productivity gap better than model quality does.
- The gains are individual, not organizational. Microsoft’s 2026 Work Trend Index found organizational factors drive 67% of AI’s measured impact versus 32% for individual ones — yet only 13% of workers say AI has meaningfully improved their organization’s performance.
- The original claim here: AI doesn’t cause coordination neglect — it amplifies it. Every agent you add is another division of labor, and division was never the thing teams were bad at. Integration was.
- New 2026 research on shared-workspace human–AI teams found that adding a capable collaborator can lower performance when the team has no structure to coordinate contributions — and that shared memory plus human approval gates reverse it.
- The fix isn’t a smarter model or another chat window. It’s a shared board where integration is designed in: agents claim and ship work in the open, status is verifiable, and nothing counts as done until it’s been brought back together.
What is coordination neglect, and why does it matter for AI teams in 2026?
Coordination neglect is the tendency to underestimate the work of integrating what a team produces. In their foundational paper, “Coordination Neglect: How Lay Theories of Organizing Complicate Coordination in Organizations” (Research in Organizational Behavior, 2000), Heath and Staudenmayer showed that to organize anything, you have to do two things: divide the task into parts, then integrate the parts back into a whole. People are intuitively good at the first and reliably blind to the second. They diagnosed two habits behind it — partition focus, obsessing over how to split the work, and component focus, fixating on individual pieces instead of the seams between them. Cross-functional teams fail, they argued, far more often from coordination breakdowns than from a lack of effort or talent.
Hold that up against how most companies are adopting AI, and the fit is uncomfortable. The entire pitch of agentic AI is division of labor at a new scale: hand a slice of the work to a model, then another slice to another model. That’s partition focus with a budget. The integration — making the outputs consistent, reconciling where they conflict, deciding what’s actually finished — is exactly the part a 2000-vintage bias predicts we’ll skip. And skipping it is cheap right up until it isn’t. We’ve written about the coordination bottleneck that replaced the coding one; coordination neglect is the psychology underneath it.
Why doesn’t AI productivity reach the whole organization?
Because individual speed and organizational output are different variables, and only the first one is going up. Microsoft’s 2026 Work Trend Index put a number on it: organizational factors account for 67% of AI’s measured impact — more than double the 32% attributable to individual mindset and behavior. Active agents in its ecosystem grew 15x year over year, and 18x in large enterprises. But only 19% of AI users sit in what Microsoft calls the “Frontier” zone, where individual capability and organizational readiness actually reinforce each other. The capability is everywhere. The readiness to integrate it is rare.
The lived version of that gap is starker. Glean’s Work AI Index 2026 found 75% of workers say AI makes them more productive and that automation alone gives back roughly eleven hours a week — yet just 13% say AI has significantly improved their organization’s performance and outcomes. Eleven hours saved per person, and almost nothing showing up at the top line. That’s not a productivity paradox so much as a coordination one: the time is real, but it’s being reabsorbed into the overhead of making disconnected AI output fit together. We unpacked the same divide in the AI productivity paradox — individual gains that never become team gains.
Does adding AI agents to a team actually make it better?
Not automatically — and there’s now direct evidence it can make things worse. A June 2026 study, “Searching for Synergy in Shared Workspace Human–AI Collaboration”, tested exactly the setup enterprises are racing toward: humans and AI agents sharing a workspace, dividing responsibilities, submitting a combined result. Its headline finding is one every team lead should sit with — adding a relevant, capable collaborator can genuinely lower a team’s performance when the team lacks the structure to coordinate contributions. The researchers frame it in exactly these terms: coordination theory treats collaboration as managing dependencies between activities, and coordination neglect is what happens when a team underweights that integration work. More hands, less output. It’s the bias, reproduced in silicon.
The market is already pricing this in. Gartner predicts that more than 40% of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls. Read that against the research and the causes rhyme: these aren’t projects that failed because the agent couldn’t do its slice. They’re projects where nobody owned the seams. The same pattern shows up when you scale agents up — more of them, working in sequence, can degrade results rather than improve them. Capability was never the ceiling. Coordination is.
The original take: AI amplifies coordination neglect — it doesn’t remove it
Here’s the synthesis, and it’s the whole argument. Coordination neglect was always an imbalance between two forces: division of labor, which humans do eagerly, and integration, which they avoid. For twenty-five years that imbalance was bounded by a simple constraint — you could only divide work as far as you had people to divide it among. Agentic AI removes the constraint. You can now spin up a near-unlimited number of workers, each taking a slice, in an afternoon. That is a massive, one-sided boost to the exact force the bias already over-weighted. The integration side — the human judgment about how it fits, what conflicts, what’s truly done — gets no such boost. So the ratio that defines coordination neglect doesn’t improve with AI. It gets dramatically worse.
Which means the instinctive fix is the wrong one. Faced with disconnected AI output, the reflex is to add another agent to reconcile it — more division, aimed at a problem caused by too much division. The study points the other way. What restored synergy in those shared-workspace teams wasn’t a better model; it was structure: a shared group memory every participant could see, plus human-in-the-loop gates where selected actions required approval from a designated teammate before they counted. Shared context, and a point where the work has to be brought together and validated. That combination lifted mean performance the most in three-person teams — small groups of mixed humans and agents, which is what most real teams are becoming. The paper effectively describes the antidote to coordination neglect. It just doesn’t call it a product.
How does a shared board fix coordination neglect for human–AI teams?
By turning integration from an act of memory into a property of the workspace. A shared board is the structure the research says works, made operational: shared group memory becomes a board every agent and human reads from and writes to, so no one is coordinating from a private context. Human-in-the-loop gates become explicit status transitions — a task can’t jump to done without passing through a state where the work is reconciled and verified. The seams that coordination neglect makes invisible are rendered as columns, owners, and checks that something has to move through. Nobody has to remember to integrate, because the board won’t let work finish until it’s integrated.
This is what Lova is built to be. An AI agent on a Lova board is a first-class teammate — it claims a task under its own identity, works in the same shared context as everyone else, and advances the task through states the whole team can inspect, attaching the evidence that “done” is really done. You can add ten agents or a hundred; the board is the one place their work is forced back together, with a human gate wherever judgment is needed. The bias that’s quietly canceling 40% of agentic projects doesn’t get a foothold, because the thing it preys on — neglected integration — is the thing the board makes unavoidable. Heath and Staudenmayer told us in 2000 where teams break. In 2026, with AI multiplying the division of labor by the day, building the integration in is no longer good hygiene. It’s the difference between an AI team that ships and one that stalls.
Frequently asked questions
What is coordination neglect?
Coordination neglect is a bias, named by Chip Heath and Nancy Staudenmayer in 2000, in which teams underinvest in integrating their work relative to dividing it up. Organizing anything requires both dividing a task into parts and integrating those parts into a whole; people are intuitively good at division and reliably neglect integration, which is why cross-functional teams tend to fail at the seams rather than inside any single component.
Why doesn’t AI improve organizational performance if individuals are more productive?
Because individual speed and organizational output are different measures. Microsoft’s 2026 Work Trend Index found organizational factors drive 67% of AI’s impact versus 32% for individual ones, and Glean’s Work AI Index 2026 found that while 75% of workers feel more productive, only 13% say AI has significantly improved their organization’s outcomes. The saved time is real but gets reabsorbed into the overhead of making disconnected AI output fit together — a coordination problem, not a capability one.
Can adding more AI agents make a team worse?
Yes. A June 2026 study of shared-workspace human–AI teams found that adding a capable collaborator can lower performance when the team has no structure to coordinate contributions, and Gartner predicts more than 40% of agentic AI projects will be canceled by the end of 2027. Adding agents increases the division of labor without adding integration, so without a shared surface to reconcile the work, more agents can mean less output.
How do you prevent coordination neglect on an AI team?
Give the team structure for integration, not just division. Research points to two things: shared memory every participant can see, and human-in-the-loop gates where work has to be approved before it counts. A shared project board operationalizes both — a common context plus explicit status transitions that force work to be reconciled and verified before it’s marked done.
What is Lova?
Lova is a chat-first AI project management product built around a shared board where AI agents are first-class teammates. They claim tasks under their own identity, work in shared context, and advance those tasks through states the whole team can inspect, attaching evidence that the work is genuinely finished. By making integration a property of the board rather than an act of memory, Lova is designed to keep coordination neglect from eating the gains AI creates.