There is a point at which many generative AI projects begin to lose momentum. It rarely happens because the technology does not work. More often, it comes a few weeks after kick-off, when the initial enthusiasm gives way to a more fundamental question: how can we turn AI’s potential into tangible, sustainable results that people across the organization can contribute to?
The challenge is not primarily technological, but methodological. According to S&P Global Market Intelligence, which surveyed more than 1,000 companies across North America and Europe, in 2025, 42% of organizations reported abandoning most of their AI initiatives, more than double the 2024 figure.
This is where Agile can make a difference. Introduced through the Agile Manifesto in 2001, Agile replaces rigid, sequential planning with short development cycles, continuous feedback, and close collaboration. It helps organizations learn and adapt as they move forward and, crucially, it is as much about people as it is about processes.
When Uncertainty Is the Rule, Not the Exception
Generative AI projects are inherently exploratory. Even with the right expertise in place, it is difficult to know at the outset exactly where they will lead. Organizations begin with a hypothesis, test it, gather feedback, refine the underlying data, and adjust course as they learn.
Traditional project management works well when the destination is clear. With generative AI, it often is not. The challenge is therefore not to plan more, but to learn faster.
This is precisely the kind of uncertainty Agile is designed to manage. Rather than defining everything in advance and then executing against a fixed plan, teams move forward in stages, learn from the results, and use those insights to determine what comes next.
The Minimum Viable Product as an Organizational Principle
Agile works through short, iterative, and measurable increments. Rather than designing an entire system before putting it into use, teams build something functional within a few weeks, typically a minimum viable product (MVP), test it with real users, gather feedback, and decide how to move forward.
This fundamentally changes the role of failure. Discovering after six months that a solution does not meet users’ needs can be costly. Discovering it after three weeks provides valuable information while there is still time to change direction. Mistakes become part of the learning process rather than its endpoint.
Companies such as Amazon and Spotify have long applied this principle, releasing features, observing user behavior, and refining them based on what they learn. The objective is not to eliminate uncertainty before taking action, but to learn quickly enough to navigate it effectively.
Agile Is About Teams Before Processes
Agile is sometimes reduced to sprints, boards, and recurring meetings, but its real value goes deeper. At its core, Agile creates the conditions for teams to work, learn, and solve problems effectively together. The Agile Manifesto itself reflects this principle by placing individuals and interactions ahead of processes and tools.
Sprint planning, daily stand-ups, reviews, and retrospectives should therefore be more than procedural routines. At their best, they create structured opportunities for dialogue, bringing together people with different expertise and perspectives to examine what has been built, identify what is working, and discuss what needs to change.
A methodology cannot generate better ideas on its own. What it can do is create an environment in which people develop better ideas together.
AI Performs Best When Used by a Team
This collaborative dimension is particularly relevant to generative AI. A 2025 study by researchers from Harvard and Wharton, conducted in collaboration with Procter & Gamble, examined how 776 professionals approached real-world product innovation challenges. In The Cybernetic Teammate, participants were randomly assigned to work individually or in teams, with or without AI.
One finding attracted considerable attention: individuals using AI achieved performance comparable to teams working without it. At first glance, this could suggest that AI reduces the need for teams. Another finding, however, tells a more interesting story: the highest-quality solutions, those in the top 10%, were produced most frequently by teams working with AI.
The distinction matters: AI improves individual performance, but AI combined with human collaboration can achieve more. It also helped bridge functional silos, with technical professionals becoming more commercially oriented and commercial professionals producing more technically robust ideas.
This points to a broader lesson for organizations. When AI is treated simply as an individual productivity tool, it risks reducing interaction and narrowing the exchange of ideas. Integrated into collaborative work, it can have the opposite effect, strengthening dialogue, constructive disagreement, and the collective development of ideas.
The most effective model may therefore be neither people working without AI nor AI working in place of people, but people and AI working together.
From Strategic Exploration to Execution
Organizations approaching generative AI typically need to navigate two distinct phases, each requiring a different mindset.
The first is strategic exploration. Before deciding what to build, organizations need to understand what they actually want to achieve with AI, which processes are worth transforming, and which business objectives should take priority. At this stage, uncertainty creates space to challenge assumptions, understand needs, and identify the opportunities most likely to generate value.
This is where approaches such as Design Thinking can play an important role. Moving too quickly through this phase creates a familiar risk: developing technically impressive solutions to problems that were never genuine business priorities.
The second phase is execution, when the opportunity is translated into something tangible, tested in practice, and improved over time. This is where Agile comes into its own. Short cycles and continuous feedback turn change from a disruption to the plan into useful information. With every iteration, both the solution and the team gain knowledge that was not available at the outset.
The Real Risk Isn’t Making Mistakes, but Failing to Learn
Agile practices are now widespread across industries, but adopting Agile terminology and developing genuine Agile capabilities are very different things. The important distinction is not between organizations that know Agile and those that do not, but between those that treat it as a set of processes and those that embrace it as a way of working and as a team culture.
For generative AI initiatives, this distinction has practical consequences. Investment in technology alone is not enough. Without the right methodology, organizational environment, and level of collaboration, even technically sound projects can struggle to translate their potential into lasting business value.
As technology evolves faster than plans, processes, and skills, competitive advantage will increasingly depend on an organization’s ability to learn and adapt. Organizations learn through people who test ideas, challenge assumptions, share perspectives, and make sense of change together.
Twenty-five centuries ago, in Alcibiades I, Plato has Socrates suggest that we come to know ourselves by seeing ourselves reflected in another person. There is a modern organizational truth in that idea: we often think more clearly when we engage with someone who understands our perspective but sees the problem differently enough to challenge it.
Agile turns that exchange into a way of working. AI can add another perspective to the conversation, another mirror around the table. But if we want AI to make organizations more intelligent rather than simply more automated, the most important reflections will still come from the people around it.
