There is a phase transition happening inside knowledge-work teams right now, and most managers are misreading it. They see the productivity numbers go up, the output improve, the meeting time shrink \u2014 and they assume the team has gotten better. Sometimes that’s true. Sometimes the team has gotten more brittle.<\/p>\n
The difference turns on a concept from complexity science: the edge of chaos.<\/p>\n
Stuart Kauffman didn’t coin “the edge of chaos” as a metaphor. He was describing a measurable structural property of complex adaptive systems \u2014 biological, economic, and organizational. In his work on NK fitness landscapes, Kauffman found that systems capable of rich adaptation and genuine creative problem-solving aren’t maximally ordered and aren’t maximally disordered. They inhabit a narrow transition zone between the two: enough structure to coordinate, enough variability to evolve.<\/p>\n
Too far toward order, and the system freezes. It executes well on known problems and fails on novel ones. Too far toward disorder, and it can’t accumulate learning \u2014 every perturbation sends it spinning with no recovery path. The edge of chaos is where the action is. It’s where evolution happens, where organizations learn, where teams generate genuinely new solutions rather than recombining old ones.<\/p>\n
The traditional high-performing team sat relatively close to the frozen end. Predictable processes. Defined roles. Clear escalation paths. That’s not a flaw \u2014 it’s how you build reliability at scale. But it also means the team is optimized for executing the known, not for navigating the novel. Kauffman would call it a fit local maximum: comfortable, hard to leave, and increasingly vulnerable as the environment shifts.<\/p>\n
AI is doing something specific to this equilibrium, and the data is striking.<\/p>\n
Harvard Business School researchers studying 791 professionals at Procter & Gamble found that individuals working with AI delivered nearly 40% performance gains, elevating them to match the output of traditional cross-functional teams. But here’s the more interesting result: AI-augmented cross-functional teams didn’t just improve \u2014 they were three times more likely to produce the top 10% of solutions. Positive emotions among AI-assisted team members rose 64%, well above what individuals or conventional teams showed.<\/p>\n
Something is genuinely changing. Not just speed \u2014 the character of the output is different. The variation in ideas is higher, the creative range wider. From a complexity lens, this is exactly what you’d expect from a system that has been moved toward the edge. AI injects cognitive slack: it handles retrieval, synthesis, and variation-generation at a rate human working memory can’t match, freeing the team to do what complexity theorists call “higher-order recombination.” The team isn’t working harder; it’s operating at a different point on the order-disorder spectrum.<\/p>\n
For organizations that were over-rigid, this is the shift that good management always aimed for and never quite achieved. AI is doing what good managers have always tried to do by less efficient means: introduce productive tension, surface unexpected combinations, reduce the friction that prevents adaptive behavior.<\/p>\n
But “toward the edge” isn’t the same as “at the edge.” You can overshoot.<\/p>\n
The first signal is automation complacency. Research on AI reliance across professional contexts shows a consistent pattern: heavy users of AI tools display measurably reduced critical thinking. They over-defer to AI-generated recommendations even when those recommendations are wrong. In medical diagnostics, in legal research, in financial analysis \u2014 the same dynamic appears. Professionals who routinely offload judgment to AI systems stop applying the scrutiny that made their judgment valuable.<\/p>\n
This matters for teams because team capability isn’t just the sum of individual skills. It’s the interplay between them: the pushback, the challenge, the cross-verification that catches errors before they propagate. When each member of a team has offloaded the same class of judgment to the same AI tool, that distributed error-checking function disappears. The team looks high-performing until it encounters a situation the AI handles badly, and then there’s no internal redundancy to catch the failure.<\/p>\n
The second signal is coordination breakdown. A 2025 longitudinal study by Qing Xiao and colleagues at Carnegie Mellon, MIT, Stanford, and Emory \u2014 “AI Hasn’t Fixed Teamwork, But It Shifted Collaborative Culture” (arXiv:2509.10956) \u2014 tracked members of a project-based software development organization from 2023 to 2025 and found exactly this: AI tools sped up individual tasks substantially, but the core collaboration challenges within teams persisted. Who knows what. Who decided what. How the team maintains a shared understanding of the problem it’s solving.<\/p>\n
This is the coordination paradox of AI-augmented teams. Individual throughput rises. But the team’s shared mental model \u2014 the collective representation of the problem, the constraints, the state of play \u2014 can fragment, because each person is moving faster through their individual slice of the work without the synchronization friction that once maintained coherence. When novel situations arise or the AI tool misbehaves, the team may find that it no longer has the internal resilience to recover.<\/p>\n
This is what Kauffman would call decohering from the edge. The system had achieved adaptive complexity, then retreated. Not to the ordered stability it started from, but to a new kind of rigidity: dependency on external tools, loss of internal cognitive variety, reduced capacity to self-organize under stress. It looks dynamic from the outside. It’s actually fragile.<\/p>\n
The question is not whether AI belongs in your team. It does. The question is whether your team is learning to navigate toward the edge or drifting past it.<\/p>\n
Three things distinguish teams that stay on the edge from those that overshoot it.<\/p>\n
The first is preserving human deliberation loops. Some organizations have begun running structured “AI-free” problem-solving sessions \u2014 not as anti-AI theater, but as deliberate practice for the reasoning muscles that AI use tends to atrophy. This isn’t about occasionally going without a tool. It’s about maintaining the team’s internal repertoire of responses, the adaptive buffering that makes it resilient when the external tool fails or gives bad output.<\/p>\n
The second is deploying AI asymmetrically. AI’s power is in generating variation quickly: scanning option space, synthesizing existing knowledge, producing drafts. Its weakness is in final judgment on novel, high-stakes problems \u2014 precisely where diverse human perspectives and genuine disagreement are most valuable. High-performing AI-augmented teams tend to use AI heavily in divergent phases and keep humans firmly in the loop during convergent ones.<\/p>\n
The third is building team meta-awareness. Teams need to track not just what they produce but how they think. Is the team reasoning through a hard problem or delegating it? Is disagreement happening because people have different views, or has AI-mediated consensus silenced the productive friction? These are not soft questions. They are indicators of where the team sits on the order-disorder spectrum.<\/p>\n
BCG’s research found that 72% of organizations have adopted AI in at least one function. Only 26% have scaled it to generate significant returns. That gap is not a technology problem. The technology is accessible, affordable, and capable. It’s a complexity-management problem. Most organizations are either deploying AI in ways that don’t move them toward the edge \u2014 bolting it onto frozen processes and measuring speed \u2014 or deploying it in ways that send them past it, trading resilience for throughput.<\/p>\n
The teams that will outperform over the next decade are not the ones that adopted AI first. They’re the ones that learned to surf the edge \u2014 generating the variability and adaptive capacity that AI enables, while maintaining the internal coherence that makes genuine adaptation possible. That requires understanding what you’re actually managing: not a technology rollout, but the complex dynamics of a system in transition.<\/p>\n
Kauffman’s insight was that the edge of chaos is not a destination. It’s a posture. You have to keep working to stay there.<\/p>\n","protected":false},"excerpt":{"rendered":"
How AI moves teams to adaptive complexity\u2014and when it tips into fragility. Understanding Kauffman’s edge of chaos in knowledge work.<\/p>\n","protected":false},"author":4,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"nf_dc_page":"","om_disable_all_campaigns":false,"_monsterinsights_skip_tracking":false,"_monsterinsights_sitenote_active":false,"_monsterinsights_sitenote_note":"","_monsterinsights_sitenote_category":0,"_jetpack_newsletter_access":"","_jetpack_dont_email_post_to_subs":false,"_jetpack_newsletter_tier_id":0,"_jetpack_memberships_contains_paywalled_content":false,"_jetpack_memberships_contains_paid_content":false,"footnotes":""},"categories":[9],"tags":[24,12,23,21,22],"class_list":["post-46","post","type-post","status-publish","format-standard","hentry","category-ai-and-society","tag-adaptive-systems","tag-artificial-intelligence","tag-complexity-science","tag-edge-of-chaos","tag-team-dynamics"],"yoast_head":"\n