Agility and AI: Another Vision on the Impact of AI on Agile Roles

Introduction: The Manufactured Crisis

Following the recent article published by our Devoteam colleagues on the impact of AI on Agility and Agile roles, we felt inspired to weigh in. While the debate often focuses on the survival of these roles, we want to share a different perspective—one that examines common misconceptions about these positions and explores the untapped opportunities AI offers to those ready to evolve.

Much of the confusion in Agile discourse is manufactured by consulting firms. They invented a crisis around “Agile roles”—categories that exist more in marketing materials than in reality—and created the abstract “Agile mindset” as a product to sell. Now, AI evangelists prophesy the end of these fabricated roles, offering solutions to problems the consulting industry invented—problems that only generate waste.

The circularity deepens: consulting firms first created expensive role structures, then sell AI solutions to eliminate those same expensive roles. They seduce clients by promising cost savings, replacing the consultants they previously insisted were essential. The industry manufactures the problem, sells the cure, then sells the antidote to its own cure.

Meanwhile, Agile’s actual principles about adaptability and delivering value get ignored because they can’t be easily packaged and sold.

The Misconception Problem: Why People Fear AI Will Replace Agile Roles

The Scrum framework was designed for simplicity and accessibility. However, this creates what might be called an “easy to play, hard to master” paradox. Many of the deeper activities and the true purpose of the roles remain hidden, misunderstood, or forgotten over time, leading to significant misconceptions about what a Product Owner or a Scrum Master actually does.

The simplification of the sixteen pages of the Scrum Guide has led to oversimplified roles, stripped of the principles behind their accountabilities. One explanation for this superficiality could be that it protects people from the cognitive dissonance between ground reality and theory—a psychological buffer that allows practitioners to function without confronting uncomfortable truths about their practice.

This leads to significant misconceptions about what Scrum roles actually entail—and by extension, how AI will impact them. The fear of AI replacement is, in most cases, a fear born from a shallow understanding of what these roles were meant to be.

Product Owner: From Backlog Clerk to Entrepreneur

Frequently, companies—and even some of our fellow agilists—frame the Product Owner as the person who “writes User Stories” and prioritises the backlog. They see the PO as a simple focal point for stakeholders. This vision often stems from a literal, surface-level reading of the Scrum Guide.

But the Scrum Guide never mentions User Stories (a common mistake), and we find other supposed accountabilities—such as validating the work of the developers or writing business acceptance criteria—that simply do not exist in the official framework. What drives genuine progression toward product ownership is a true conviction in a vision or something greater. Those who merely copy and paste practices, living in their comfort zones, must contend with cognitive dissonance—and will increasingly find themselves confronting the rise of AI as a direct challenge to their relevance.

This misconception often stems from the “copy-paste” migration, where companies rebranded Project Managers as Product Owners without changing the underlying mindset—thinking it was essentially the same work but with the accountabilities listed in the Scrum Guide tacked on. Frameworks like SAFe, with their dual operating system concept and the splitting of accountabilities between Product Owner and Product Manager roles, have not helped clarify matters. As Lise Davies observes in her work on Agile coaching, “transformation fails when organisations change titles without changing thinking”.

What Is a Real Product Owner?

But a real Product Owner is so much more. Jeff Sutherland, the co-creator of Scrum, describes the PO as someone who can persuade without formal managerial authority.  They are the ones who:

• Observe and study the outside world and market trends
• Orient the product by constantly adapting the vision
• Decide by creating new business hypotheses
• Act by testing these hypotheses through the team’s work

You can find in the resources at Scrum.org a graphic illustrating how a Product Owner—a real one, a good one—with time becomes an Entrepreneur. 

• They are the ones who study the market, the needs of the customers, and all data that can be useful. 
• They look at the cost of investment and the revenue hypothesis for each functionality.
• They decide whether to spend money on a feature based on technical expert advice, development costs, and market studies.

As the Scrum Guide explicitly states: “the entire organisation must respect their decisions”. The Product Owner is the one who calls the shots. It is all in the name: Product Owner. They own the product. And while they may follow high-level objectives, KPIs, or OKRs coming from leadership, they remain the one who creates and decides on the vision of how and what to do to achieve those objectives. They are not a middleman; they are the strategist of their own scope.

Why This Matters in the Age of AI

If we look at the “classical” PO—the Scribe or the Proxy—a vast majority of their activities can, and should, be augmented or replaced by AI. An AI agent can already write excellent User Stories from specifications. It can prioritise a backlog based on complexity and projected impact. It can automate business acceptance criteria tests and generate KPI reports.

And that’s okay. These tasks are not the heart of the role.

The true impact of a Product Owner lies in exploring user needs—using frameworks like Jobs to be Done —and making the “good call” on what to try and what not to try. Being an entrepreneur is something AI cannot replicate. It can gather data and calculate statistics, but it cannot make the final “bet.” A product decision is a synthesis of data-driven projection and informed intuition. It is a hypothesis about the future, not a mathematical certainty.

It requires a form of managerial courage, much like the courage it takes to delete useless meetings from an overcrowded agenda. Any waste that can be automated will be removed over time.

The verdict is clear: If you remain a “Scribe” or a “Proxy,” AI will likely take your job. But if you step up into an Entrepreneurial posture, AI becomes your best friend. It frees you from clerical overhead so you can focus on what truly matters: creating value.

Scrum Master: Beyond the Ceremony Animator

If the misconception of the Product Owner role is common, the misunderstanding of the Scrum Master is perhaps even deeper and more difficult to define.

In many companies, the role is reduced to a “Meeting Animator” or a “Ticket Guardian.” We’ve all seen these caricatures:

The “Club Med” GO (Gentle Organiser): They schedule ceremonies, take notes, and ensure everyone has their post-its.

The Jira Cop (or Ticket Master): They police the board, hunt down missing updates, and build dashboards to track the team’s every move.

The Happiness Manager: They focus solely on keeping the team “happy” with fun retrospectives, often forgetting the goal of performance.

We can’t say this comes from a surface reading of the Scrum Guide, as the Guide itself lists many accountabilities that have nothing to do with the roles mentioned above. Rather than expanding on the Guide, I’ll propose another idea: The Scrum Master is an Agile Coach.

If we take this hypothesis, let’s examine the postures listed by Lyssa Adkins in Coaching Agile Teams and see how AI impacts each one.

Animation

Animation with intent is about selecting or creating the right structure and agenda for a meeting, depending on the context and objectives. This is where we usually find our “Club Med” GO or our Happiness Manager. AI agents can already craft excellent agendas or structures if you explain your needs and context. It can even give you directives on how to act. A professional is still a plus—for now—because of their experience in adapting in real time, but that is becoming a “comfort” feature. If you only facilitate structure, it will be complicated to justify your value. However, if you use AI to help you craft better structures, merging your knowledge with its speed, it becomes a great tool to win back time for other postures. Interactions between humans are easier when facilitated by another human.

Teacher

This involves teaching the framework, a workshop, or a tool to help the team create a better environment. It’s where you explain the “Why” and the “How” of simplifying procedures. The impact of AI here is mixed. It can explain concepts well (though some companies, like SAFe, are now protecting their content against AI deep learning), but it misses the subtleties. Most importantly, AI cannot be in the room observing interactions or analysing what works in real time. For now, you have a step ahead as long as you can propose adaptations based on what you see. But if you only “preach” the manual, AI will eventually beat you at that game.

Coach-Mentor

Mentoring is about providing possibilities, resources, or schools of thought when someone asks for help. Coaching is helping a person find their own answer by asking the right questions. If you know how to customise your AI agent, it can be very effective in these two postures. These roles demand a vast knowledge of tools or specific question protocols—two things AI is very good at. Many agilists already use AI as a “sparring partner” to confront their ideas.

However, it requires you to take a step back. Like with a human, not all answers are good; it’s not about facts, and you must still determine the good from the bad. An assumed Coach-mentor stance as an Agile Coach is to provide guidance when the coachee goes against the values of Agile and Lean. On the other side, going with the flow is far easier for an AI, as surveys show, many people’s first use of AI resembles consulting a psychologist.

Conflict Navigator

This is where you analyse a conflict, decide whether to intervene, and teach the team how to navigate it themselves. This is where the true “soft skills” arise. AI agents cannot act here. It would need to store all verbal and non-verbal cues from previous conversations and analyse team interactions in real time. Furthermore, conflict navigation is not a hard science; it involves experiments and bets—a game AI is not good at.

Collaboration Conductor

The goal here is to help the team move from cooperation to true collaboration. In a nutshell, it’s helping the team move from a state where the work is the “sum of individual parts” to a state where the result is a synergy that cannot be traced back to an individual. It’s where you create small steps, like proposing to stop pre-assigning items on Jira or ensuring that ideas bounce from each other instead of against each other. For all the same reasons as with conflict navigation, AI agents are not currently able to execute this well.

The codified parts of the work—Facilitating and Teaching—can be replaced by an AI Agent. The “Club Med” GO, the Happiness Manager, and the Jira Cop are all replaceable because they mostly execute rules and analyse codified situations.

And that’s okay. It’s not the heart of the job.

The Risk of Agile Model Collapse

There is a warning we must raise. In the world of AI-generated images, we speak of “Model Collapse” when an AI learns from other AI-generated content, eventually producing distorted, soulless results (Shumailov et al., 2024).

If companies replace the “soul” of the Scrum Master with AI-generated rituals, we risk an Agile Model Collapse. We will end up with “Cargo Cult” teams: they will execute the ceremonies perfectly, copy-pasted by an AI, without ever understanding why they do them. They will perform the movements of agility without the spirit of it.

Which Product Are You Selling?

Ultimately, your role’s survival depends on your value proposition. Imagine your job as a product you sell to your company:

Product A: “I manage the Scrum framework and its ceremonies.” If this is your value, AI is your direct competitor, and it’s cheaper and faster than you.

Product B: “I help this team build its own best environment so they can eventually grow and adapt without me.” If this is your value, AI is your best friend. It will handle the “clerical” work—designing workshop, scheduling, reporting—freeing you to focus on the invisible: the culture, the system, and the human growth.

As for the “Jira Cops” and “Ticket Masters”? They shouldn’t have existed in the first place. AI isn’t killing those roles; it’s finally allowing us to abandon them where they belong: in the past.

The Adoption Paradox: Who’s Actually Using AI?

Walk into any executive committee discussing AI strategy and you will notice something peculiar: the people deciding how the organisation should adopt AI are often the last ones who would adopt it themselves. This is not a coincidence. It is structural.

Geoffrey Moore’s Crossing the Chasm describes the Late Majority as adopting technology “only after something has become an established standard,” driven by “discomfort with technology” and a need for “lots of support”. The skills that propel someone through corporate hierarchies—consensus-building, risk mitigation, procedural mastery—are precisely those that slow the adoption of disruptive technologies.

Recent data reveals the depth of this disconnect. Leadership teams overestimate employee enthusiasm for AI adoption by more than two-to-one—perceiving widespread excitement where cautious ambivalence actually exists (Gartenberg, 2025). Meanwhile, these same executives recognise AI as strategically important yet struggle to articulate long-term transformation visions, defaulting instead to governance frameworks and vendor evaluations (McKendrick, 2025). Research comparing AI systems to human executives in simulated strategic scenarios shows AI excelling at data-driven optimisation but failing catastrophically at navigating unexpected disruptions—revealing the exact inverse of executive strengths (Puranam & Vanneste, 2024). The irony is striking: those least equipped for AI adoption lead its deployment, while those most capable of leveraging it wait for permission.

Zones of Uncertainty

Philippe Bernoux’s Sociologie des organisations explains this. Organisations contain “zones of uncertainty”—informal spaces where official rules have not reached. Employees exploit these gaps not out of subversion but necessity. “The actor rarely has clear objectives,” Bernoux writes. “He discovers them through action”. The analyst did not plan to revolutionise document workflow; she needed to finish a report by Thursday.

The consequences are predictable. A multinational launches a six-month AI governance initiative. Steering committees form. Vendors are evaluated. Meanwhile, a junior analyst has been using Claude to draft client reports since February. A project manager quietly runs meeting summaries through ChatGPT. By the time the official “approved tools list” circulates, employees have already built sophisticated workflows. Research suggests that a significant majority of knowledge workers now use AI tools daily, often without organisational sanction.

What emerges is a pattern familiar from other organisational transformations: genuine innovation happens at the edges, where people closest to actual work discover what tools and practices actually function. These frontline workers develop solutions that reflect operational reality rather than strategic aspiration. Their experiments are small, reversible, and tied to immediate problems. When something works, it spreads laterally through informal networks long before appearing in official documentation.

So when executives publish “Which AI Tool Should Your Team Use?”—they are documenting archaeology—describing practices already established, workflows already optimised, value already captured. The question was answered months ago, in practice, by people with dirty hands. The real opportunity is not to prescribe but to learn from how individuals are already interacting with AI.

Beyond the X10 Obsession: AI as Structured Difference

Open any AI productivity thread, and you will drown in acceleration metaphors. “10x your writing.” “Ship code in half the time.” This framing shapes how organisations justify AI investment: time saved, output increased,and  headcount implications.

But the fixation on speed obscures something more valuable—and more complicated.

Consider what happens when a team uses AI well. A designer asks Claude to argue against her proposed solution. A strategist generates twelve market approaches instead of three. A researcher surfaces contradictions in her own thinking. These are not acceleration use cases—they are expansion use cases. Yet the research on human-AI collaboration tells a more nuanced story than early enthusiasm suggested.

A meta-analysis of 106 studies found that human-AI combinations, on average, performed worse than the best of humans or AI alone—with losses in decision-making tasks but gains in content creation (Vaccaro et al., 2024). Meanwhile, when multiple people brainstorm with ChatGPT, their ideas converge: what individuals gain in quality, collectives lose in diversity. In one study, 94% of AI-assisted ideas shared overlapping concepts; nine participants independently named their toy “Build-a-Breeze Castle” (Meincke et al., 2025). The original researchers responded by noting that real brainstorming rarely involves everyone consulting the same AI in isolation—and that intentional prompting for divergence can recover variety (Lee & Chung, 2025). Yet in medical diagnosis, hybrid collectives consistently outperform both human-only and AI-only groups—not because AI amplifies human thinking, but because humans and AI make different errors that rescue each other (Zöller et al., 2025).

Succeeding Through Error Complementarity

The pattern emerging from this research is not “human-AI beats either alone” but something more specific: human-AI collaboration succeeds through error complementarity—when the human and AI fail differently, and each catches what the other misses.

Arnold Mindell’s Sitting in the Fire offers a framework for understanding why this matters for teams. He introduced “deep democracy”—the radical inclusion of all voices, especially those we would rather not hear. Every group contains “ghosts”: absent perspectives, silenced dissent, viewpoints that shape dynamics without being named. “The ghost role is a role that’s needed in the group but nobody is representing it”.

AI, used with intention, can make the ghosts speak—not because it is wiser than humans, but because it is differently blind. The same systematic tendencies that make AI-assisted brainstorming converge can be turned to advantage: AI will articulate positions that humans instinctively suppress.

Space for Difficult Conversations

Practitioners are experimenting concretely. In planning sessions, an AI represents the frustrated customer who churned—articulating objections the team unconsciously avoids. In retrospectives, an AI surfaces what junior members think but will not say. Mindell warned that groups instinctively silence uncomfortable voices: “Most of us have been trained to marginalise and oppress aspects of our experience that do not fit with the norms of mainstream culture.” AI creates a container for those voices without the social risk a human would face. The value is not that AI thinks better, but that it thinks otherly—and in that difference lies the complementarity the research identifies.

This application aligns with what Lise Davies describes as the coach’s role in “holding space for difficult conversations”. The AI becomes not a replacement for human facilitation but a tool that extends the coach’s capacity to surface what would otherwise remain hidden.

The X10 narrative keeps our eyes on the speedometer. But speed was never the transformation. The transformation is structured difference—designing interactions where AI’s systematic blindspots meet human correction, and human systematic blindspots meet AI challenge. Not doing the same things faster, but failing differently together.

Conclusion: Surfing the Wave

The threat AI poses to Agile roles is real—but only for those who have reduced their practice to its most superficial expressions. For Product Owners who have become backlog clerks, for Scrum Masters who have become ticket administrators or Scrum Enforcer, AI represents an existential challenge. But for those who embrace the entrepreneurial and coaching dimensions of these roles—the dimensions that were always intended—AI becomes a force multiplier.

The Agile Manifesto called for “continuous attention to technical excellence and good design”. Today, that excellence includes understanding how to leverage AI not merely for efficiency but for the deeper work of creating value, facilitating genuine collaboration, and surfacing the perspectives that organisations most need to hear.

The wave is coming. The question is not whether you will be swept away, but whether you will learn to surf.

---

References

• Adkins, L. (2010). Coaching Agile Teams: A Companion for ScrumMasters, Agile Coaches, and Project Managers in Transition. Addison-Wesley Professional.
• Beck, K., Beedle, M., van Bennekum, A., Cockburn, A., Cunningham, W., Fowler, M., Grenning, J., Highsmith, J., Hunt, A., Jeffries, R., Kern, J., Marick, B., Martin, R. C., Mellor, S., Schwaber, K., Sutherland, J., & Thomas, D. (2001). Manifesto for Agile Software Development. https://agilemanifesto.org/
• Bernoux, P. (2015). Sociologie des organisations: Initiation théorique suivie de douze cas pratiques (7th ed.). Éditions du Seuil.
• Christensen, C. M., Hall, T., Dillon, K., & Duncan, D. S. (2016). Know your customers’ “Jobs to Be Done”. Harvard Business Review, 94(9), 54–62.
• Davies, L. (2022). The Agile Coach’s Guide to the Galaxy: Working with Agile Teams. Agile Alliance Publications.
• Gartenberg, C. (2025, November 17). Leaders assume employees are excited about AI. They’re wrong. Harvard Business Review. https://hbr.org/2025/11/leaders-assume-employees-are-excited-about-ai-theyre-wrong
• Lee, B. C., & Chung, J. J. (2024). An empirical investigation of the impact of ChatGPT on creativity. Nature Human Behaviour, 8, 1906–1914.
• Lee, B. C., & Chung, J. J. (2025). Reply to: ChatGPT decreases idea diversity in brainstorming. Nature Human Behaviour, 9, 1110–1112.
• McKendrick, J. (2025, January 27). Too many executives sweat the small stuff with AI versus long-term value. Forbes. https://www.forbes.com/sites/joemckendrick/2025/01/27/too-many-executives-sweat-the-small-stuff-with-ai-versus-long-term-value/
• Meincke, L., Nave, G., & Terwiesch, C. (2025). ChatGPT decreases idea diversity in brainstorming. Nature Human Behaviour, 9, 1107–1109.
• Mindell, A. (1995). Sitting in the Fire: Large Group Transformation Using Conflict and Diversity. Lao Tse Press.
• Moore, G. A. (2014). Crossing the Chasm: Marketing and Selling Disruptive Products to Mainstream Customers (3rd ed.). HarperBusiness.
• Puranam, P., & Vanneste, B. (2024, September 3). AI can (mostly) outperform human CEOs. Harvard Business Review. https://hbr.org/2024/09/ai-can-mostly-outperform-human-ceos
• Schwaber, K., & Sutherland, J. (2020). The Scrum Guide: The Definitive Guide to Scrum: The Rules of the Game. https://scrumguides.org/
• Shumailov, I., Shumaylov, Z., Zhao, Y., Gal, Y., Papernot, N., & Anderson, R. (2024). AI models collapse when trained on recursively generated data. Nature, 631, 755–759.
• Sutherland, J. (2014). Scrum: The Art of Doing Twice the Work in Half the Time. Crown Business.
• Vaccaro, M., Almaatouq, A., & Malone, T. (2024). When combinations of humans and AI are useful: A systematic review and meta-analysis. Nature Human Behaviour, 8, 2293–2303. https://doi.org/10.1038/s41562-024-02024-1
• Zöller, M. A., et al. (2025). Human–AI collectives most accurately diagnose clinical vignettes. Proceedings of the National Academy of Sciences, 122(4), e2403025121.