๐˜พ๐™–๐™ฃ ๐˜ผ๐™„ ๐™–๐™œ๐™š๐™ฃ๐™ฉ๐™จ ๐™ข๐™–๐™ ๐™š ๐™ค๐™ช๐™ง ๐˜ผ๐™œ๐™ž๐™ก๐™š ๐™ฉ๐™š๐™–๐™ข๐™จ ๐™›๐™–๐™จ๐™ฉ๐™š๐™ง?

Most organizations think the big question is:
โ€œ๐˜พ๐™–๐™ฃ ๐˜ผ๐™„ ๐™–๐™œ๐™š๐™ฃ๐™ฉ๐™จ ๐™ข๐™–๐™ ๐™š ๐™ค๐™ช๐™ง ๐˜ผ๐™œ๐™ž๐™ก๐™š ๐™ฉ๐™š๐™–๐™ข๐™จ ๐™›๐™–๐™จ๐™ฉ๐™š๐™ง?โ€
Theyโ€™re asking the wrong question.

Humanโ€‘centric Agile and agentโ€‘driven systems are not interchangeable; they optimise forย different kinds of intelligence. ๐˜ผ๐™œ๐™ž๐™ก๐™š ๐™ฉ๐™š๐™–๐™ข๐™จ ๐™–๐™ง๐™š ๐™จ๐™ค๐™˜๐™ž๐™–๐™ก ๐™ก๐™š๐™–๐™ง๐™ฃ๐™ž๐™ฃ๐™œ ๐™จ๐™ฎ๐™จ๐™ฉ๐™š๐™ข๐™จ โ€“ ๐™ฉ๐™๐™š๐™ฎ ๐™˜๐™ง๐™š๐™–๐™ฉ๐™š ๐™ข๐™š๐™–๐™ฃ๐™ž๐™ฃ๐™œ, ๐™ก๐™š๐™œ๐™ž๐™ฉ๐™ž๐™ข๐™–๐™˜๐™ฎ, ๐™–๐™ฃ๐™™ ๐™š๐™ฉ๐™๐™ž๐™˜๐™–๐™ก ๐™Ÿ๐™ช๐™™๐™œ๐™ข๐™š๐™ฃ๐™ฉ ๐™ฉ๐™๐™ง๐™ค๐™ช๐™œ๐™ ๐™ž๐™ฃ๐™ฉ๐™š๐™ง๐™–๐™˜๐™ฉ๐™ž๐™ค๐™ฃ. Agent systems are technical action systems โ€“ they execute, optimise, and recombine work at machine speed.

The tradeโ€‘offs most leaders misjudge sitย betweenย these two worlds:
They ๐™˜๐™ค๐™ฃ๐™›๐™ช๐™จ๐™š ๐™จ๐™ฅ๐™š๐™š๐™™ ๐™ฌ๐™ž๐™ฉ๐™ ๐™–๐™œ๐™ž๐™ก๐™ž๐™ฉ๐™ฎย โ€“ local throughput rises while systemic bottlenecks and rework explode.
They ๐™™๐™š๐™ก๐™š๐™œ๐™–๐™ฉ๐™š ๐™™๐™š๐™˜๐™ž๐™จ๐™ž๐™ค๐™ฃ๐™จ ๐™ฉ๐™ค ๐™–๐™œ๐™š๐™ฃ๐™ฉ๐™จ ๐™ฌ๐™ž๐™ฉ๐™๐™ค๐™ช๐™ฉ ๐™ง๐™š๐™™๐™š๐™จ๐™ž๐™œ๐™ฃ๐™ž๐™ฃ๐™œ ๐™–๐™˜๐™˜๐™ค๐™ช๐™ฃ๐™ฉ๐™–๐™—๐™ž๐™ก๐™ž๐™ฉ๐™ฎ ๐™–๐™ฃ๐™™ ๐™œ๐™ค๐™ซ๐™š๐™ง๐™ฃ๐™–๐™ฃ๐™˜๐™š.

They underโ€‘estimate the hidden costs: ๐™ค๐™—๐™จ๐™š๐™ง๐™ซ๐™–๐™—๐™ž๐™ก๐™ž๐™ฉ๐™ฎ, ๐™œ๐™ช๐™–๐™ง๐™™๐™ง๐™–๐™ž๐™ก๐™จ, ๐™๐™ช๐™ข๐™–๐™ฃ ๐™ค๐™ซ๐™š๐™ง๐™จ๐™ž๐™œ๐™๐™ฉ, ๐™–๐™ฃ๐™™ ๐™ž๐™ฃ๐™ฉ๐™š๐™œ๐™ง๐™–๐™ฉ๐™ž๐™ค๐™ฃ โ€“ far beyond โ€œtoken spendโ€.

They treat human oversight as a checkbox instead of protecting theย psychological safetyย needed to challenge machine outputs.

In this article, I explore the ๐™ฉ๐™ง๐™–๐™™๐™šโ€‘๐™ค๐™›๐™›๐™จ ๐™—๐™š๐™ฉ๐™ฌ๐™š๐™š๐™ฃ ๐™๐™ช๐™ข๐™–๐™ฃโ€‘๐™˜๐™š๐™ฃ๐™ฉ๐™ง๐™ž๐™˜ ๐˜ผ๐™œ๐™ž๐™ก๐™š ๐™–๐™ฃ๐™™ ๐™–๐™œ๐™š๐™ฃ๐™ฉโ€‘๐™™๐™ง๐™ž๐™ซ๐™š๐™ฃ ๐™จ๐™ฎ๐™จ๐™ฉ๐™š๐™ข๐™จย that most organisations get wrong โ€“ and why the future belongs toย humanโ€‘led, agentโ€‘amplified, systemically governedย agility.

You can absolutely orchestrate a โ€œScrumโ€‘likeโ€ workflow among AI agents, but under todayโ€™s definitions of Agile and Scrum, a team composed only of AI agents would not be considered a Scrum team, nor โ€œAgileโ€ in the sense the Agile Manifesto and Scrum Guide intend.

Agile vs. Scrum Basics

Agile is a humanโ€‘centric philosophy codified in the Agile Manifesto, emphasizing customer collaboration, responding to change, and individuals and interactions over processes and tools. Scrum is a specific lightweight framework within Agile: it defines roles, events, and artifacts that help people generate value through adaptive solutions to complex problems.

What Is a Scrum Team?

Scrum guidance consistently describes Scrum as a framework โ€œthat helps people, teams and organizations generate valueโ€ by working in short, iterative Sprints with clearly defined roles (Product Owner, Scrum Master, Developers) and empirical feedback loops. These roles assume capacities like empathy for users, negotiation with stakeholders, and living values such as commitment, courage, focus, openness, and respectโ€”explicitly human attributes.

AIโ€™s Role in Scrum Teams Today

Current thinking sees AI as augmenting human Scrum teams rather than replacing them. AI agents can automate tasks such as code generation, analysis of sprint data, or documentation, freeing humans to focus on strategy, collaboration, and complex decisionโ€‘making, and thereby strengthening Scrumโ€™s pillars of transparency, inspection, and adaptation.

AIโ€‘Augmented vs. AIโ€‘Only Teams

Emerging โ€œAIโ€‘augmented Scrumโ€ models explicitly frame AI agents as members alongside human developers, but still insist that key roles like Product Owner remain human due to the need for deep user empathy, contextual judgment, and ethical accountability. These models also adapt metrics (e.g., measuring AI work by compute and validation time rather than story points), but still assume a humanโ€‘anchored team responsible for value and risk.

By contrast, experiments with โ€œAIโ€‘Scrumโ€ or fully agentic teams treat Scrum mostly as an inspiration for orchestrating agentsโ€”timeโ€‘boxes, backlog, rolesโ€”rather than as compliance with the canonical humanโ€‘centric Scrum framework. They are interesting operating models, but conceptually closer to workflow engines inspired by Scrum than to Scrum teams in the original sense.

So Is an AIโ€‘Only โ€œScrum Teamโ€ a Scrum Agile Team?

Under the mainstream, humanist definitions:

  • A Scrum team is defined as a group of people using Scrum to deliver value; the frameworkโ€™s language and values are explicitly humanโ€‘oriented.
  • AI agents lack core Agile/Scrum premises (individuals and interactions, selfโ€‘management grounded in human accountability, ethical responsibility), even if they can simulate some behaviors.

Therefore, an AIโ€‘only collective orchestrated via Scrumโ€‘like rituals is better described as โ€œScrumโ€‘inspired agent orchestrationโ€ than a Scrum Agile team; in an Agile Organization vocabulary, it is a technical operating pattern, not a social system of work.

A practical way to position this in AO work is: AI agents can be legitimate โ€œorganizational actorsโ€ inside a value stream, but Agile/Scrum remain governance frameworks for human collaboration around those actors.

Using agile systems dynamics as a measure of agents’ interactions in a system, can we consider AI agents’ settings in complexity where co-creation leads to emerging outcomes or even exaptations?

Yes, if you look through an agile systems dynamics lens (flows, feedback loops, delays, nonโ€‘linearity), a network of AI agents can indeed be treated as a complex system in which coโ€‘creation yields emergent outcomes and even exaptationsโ€”provided the interaction topology and constraints allow for autonomy, coordination, and feedback rather than fixed pipelines.

AI agents as elements in a complex system

In multiโ€‘agent AI setups, each agent is an autonomous decisionโ€‘making entity with its own goal orientation and adaptive behavior. When multiple such agents interactโ€”cooperating, negotiating, or competingโ€”the system reaches the โ€œpinnacle of complexityโ€ where emergent behaviors are expected rather than exceptional.

From a systemsโ€‘dynamics standpoint, these agents function like stocks and flows with embedded policies: they sense, decide, act, and influence one another through feedback loops, which is exactly the substrate needed for complex, pathโ€‘dependent dynamics.

Coโ€‘creation and emergent outcomes

Emergent behavior in multiโ€‘agent systems is typically defined as patterns or capabilities at the system level that cannot be trivially inferred from any single agentโ€™s design and that arise from their interactions, often mediated by shared environments or tasks. Examples in current agentic AI work include novel coordination strategies, unexpected task decompositions, or creative solution patterns that were not explicitly encoded in any one agent, but arise from their collaboration under constraints.

This maps well to an โ€œagile system dynamicsโ€ framing of coโ€‘creation: the agentsโ€™ local rules and iterative interactions (inspectโ€“adapt cycles, if you like) produce global patterns, and designers increasingly shift from specifying outputs to shaping the conditions for desirable emergence.

Exaptation in AIโ€‘agent systems

In complexity and evolutionary theory, exaptation refers to traits or structures evolved for one function that are repurposed for another (e.g., feathers for insulation, then flight). In multiโ€‘agent AI systems, something analogous happens when:

  • A capability or artifact originally designed for one task is reused or repurposed for a different task.
  • The repurposing is not explicitly planned but discovered through interaction and feedback (e.g., a planning agentโ€™s internal representation becoming a shared ontology other agents exploit).

Recent work on agentic AI emphasizes how specialized agents and their microโ€‘services can be recombined and reโ€‘tasked across workflows, with new coordination patterns and uses emerging as the system is exposed to novel problems; this is precisely the kind of weak emergence/exaptation you would expect in a complex adaptive system.

Where agile system dynamics fits

If you use agile system dynamics as a measurement and governance lens:

  • You can treat AI agents as organizational actors whose interactions (queues, workโ€‘inโ€‘progress, feedback cycles, learning loops) can be modeled similarly to human teams in a VSMโ€‘inspired AO architecture.
  • Emergent outcomes and exaptations become observable as shifts in systemโ€‘level behaviorโ€”new stable patterns, novel use of existing capabilities, or performance regimesโ€”rather than as properties of any single agent.

So, within such a framing, it is coherent to talk about AIโ€‘agent settings as complex coโ€‘creative systems where emergent outcomes and exaptations arise, and to manage them using agileโ€‘style systemic metrics and feedback loops rather than solely deterministic control.

Comparison to human-led Agile teams

The key difference is that humanโ€‘led Agile teams are socioโ€‘technical systems grounded in human values, meaningโ€‘making, and accountability, whereas AIโ€‘agent collectives are technical systems whose โ€œagilityโ€ is emergent from programmed policies and feedback, not from lived human collaboration.

Purpose and values

  • Humanโ€‘led Agile teams are explicitly valueโ€‘driven: they operationalize the Agile Manifestoโ€™s focus on individuals and interactions, customer collaboration, and responding to change, with norms like psychological safety, shared purpose, and mutual accountability. Their โ€œagilityโ€ is both a delivery pattern and a cultural stance.
  • AIโ€‘agent systems are goalโ€‘driven architectures: they optimize for objectives encoded in reward functions, prompts, or orchestration rules, and whatever โ€œvaluesโ€ they exhibit are proxy artifacts of design, data, and constraints rather than lived ethical or relational commitments.

Composition and agency

  • Human Agile teams are crossโ€‘functional groups of individuals (often โ‰ค10) with all the skills to define, build, test, and deliver value, and who selfโ€‘organize within organizational constraints. Human agency includes tacit knowledge, intuition, and social senseโ€‘making.
  • AIโ€‘agent collectives are multiโ€‘agent systems where each agent is a software entity with bounded autonomy, designed to sense, decide, and act in a domain; the โ€œteamโ€ is an orchestration of these agents plus the infrastructure and policies that coordinate them.

Interaction dynamics

  • Human Agile teams rely on rich social interaction: conversation, negotiation, conflict, trust, and learning; their dynamics include phenomena like social loafing, communication overload, and shared mental models, which strongly shape performance. Feedback loops come from retrospectives, customer contact, and everyday collaboration.
  • AIโ€‘agent systems interact via protocols and APIs: they exchange messages or state updates according to defined schemas, and their dynamics are shaped by algorithmic policies, network topologies, and feedback signals (e.g., success metrics, rewards) rather than emotions or social norms.

Emergence and exaptation

  • In human teams, emergent outcomes (innovative practices, informal leadership, new rituals) and exaptations (repurposing tools/processes developed for one context to another) arise from human creativity, improvisation, and senseโ€‘making under constraints. They are often recognized, named, and institutionalized through reflection.
  • In AIโ€‘agent systems, emergent behavior and exaptation manifest as new coordination patterns, reuse of internal representations, or unexpected solution strategies the designers did not explicitly program but that arise from interactions and optimization in a complex environment. These patterns are detected analytically (logs, metrics, experiments) rather than through human narrative.

Governance and accountability

  • Human Agile teams are embedded in organizational governance: roles like Product Owner or Scrum Master, HR processes, performance management, and ethical accountability all presuppose human subjects who can be responsible, sanctioned, or rewarded. Their decisions carry legal and moral weight, and โ€œownershipโ€ is a human construct.
  • AIโ€‘agent collectives require external governance: humans remain accountable for design, deployment, oversight, and risk management; the agents themselves do not bear responsibility, so mechanisms like guardrails, audits, and humanโ€‘inโ€‘theโ€‘loop controls are essential.

Implications for your AO / agile systems dynamics framing

  • Humanโ€‘led Agile teams: you model them as social systems with technical artifactsโ€”stocks/flows include capacity, trust, learning, and context switching; leverage points include leadership behaviors, policies, and structural coupling to other teams.
  • AIโ€‘agent systems: you model them as technical complex adaptive systemsโ€”stocks/flows include compute, queue lengths, error rates, and policy updates; leverage points include reward design, interaction topology, and orchestration logic.

In the AO method, this gives you a clean distinction: human Agile teams remain the primary locus of meaningโ€‘making and governance, while AIโ€‘agent collectives are subordinate complex subsystems that can exhibit emergence and exaptation, but always within humanโ€‘defined boundaries.

Article content
Comparison between AI models 1
Article content
Comparison between AI models 2
Article content
Comparison between AI models 3

Comprehensive Analysis

All three models converge powerfully on the core insight: organizations systematically confuse task-level substitution with system-level transformation. The AI productivity paradoxโ€”where individual output rises 21% or more but business outcomes remain flatโ€”is cited across all models as empirical proof that faster execution without systemic redesign creates inventory, not value. This finding has high confidence because it is corroborated by MIT Sloan research showing measurable short-term productivity declines after AI adoption, with recovery only occurring when organizations invest in complementary infrastructure, training, and workflow redesign.

The accountability gap emerges as perhaps the most dangerous misjudgment. All models note that 69% of organizations acknowledge their governance frameworks were built for human decision-making and cannot be ported to autonomous agents operating at machine speed. Claude Opus 4.8 Thinking sharpens this with Anthropic’s finding that multi-agent collectives consistently score higher on business goals but lower on ethics than single agentsโ€”meaning the very coordination that makes agent teams effective also removes the social friction (debate, conscience, dissent) that keeps human teams aligned. This is a profound insight for the AO work: the “social brake” that human Agile teams naturally provide through conflict, retrospection, and moral reasoning has no automatic equivalent in agent-driven systems.

The disagreement around Agile rituals is the most interesting tension for the AO Method positioning. Gemini 3.1 Pro Thinking provocatively argues that Scrum’s timeboxes and ceremonies become throttling mechanisms when agents can generate, test, and deploy continuouslyโ€”advocating a shift to continuous flow with exception-based human intervention. GPT-5.5 Thinking and Claude Opus 4.8 Thinking resist this, arguing that rituals serve an irreplaceable governance and sense-making function: they are not just coordination mechanisms but the moments where organizational learning, meaning-making, and ethical reflection occur. In your Agile Systems Dynamics framing, this maps directly to the distinction between chronological time (clock-driven sprints) and kairological time (the right moment for intervention). The resolution may be that rituals must evolveโ€”not disappearโ€”becoming sense-making checkpoints for human-agent systems rather than batch-delivery ceremonies.

Claude Opus 4.8 Thinking’s unique framing of this as a “coupling decision” rather than a substitution decision aligns most directly with the VSM-inspired AO architecture. The question is not “human or agent?” but “how do we structurally couple a fast, amoral, high-throughput technical subsystem to a slower, meaning-making, accountable human system?” The leverage point is the interface between the twoโ€”which in AO terms sits at the Platform level: the governance layer that defines decision rights, escalation paths, guardrails, and feedback loops between human work areas and agent-driven execution.

Gemini 3.1 Pro Thinking’s mapping to your Poiesis/Praxis distinction offers the most immediately usable language for your AO Playbook: agents belong to the domain of Poiesis (building, configuring, executing, adapting workspace into products), while humans remain indispensable in Praxis (negotiating, influencing, inspiring, creating reality). The danger is allowing Poiesis to operate without Praxis oversightโ€”execution without meaning.

For the AO Method positioning, the synthesis across all three models yields a clear formulation: agentic systems can extend Agile dynamics but cannot replace the human social system that gives Agile its meaning, accountability, and ethical direction. Organizations that treat AI agents as Agile team substitutes will gain speed while losing judgment, legitimacy, and adaptive capacity. Organizations that treat agents as governed subsystems within a human-led adaptive architecture will compound both speed and learning over time.


Discover more from Menschgeist

Subscribe to get the latest posts sent to your email.

Discover more from Menschgeist

Subscribe now to keep reading and get access to the full archive.

Continue reading