Agile Best Practices
July 3, 2026
The Human Operating System: Why 95% of AI Initiatives Fail
# AI-Native SAFe
# AI
Three experts at SAFe Summit 2025 explain why organizational learning velocity—not better models—determines who wins in the age of AI.
Jason Flynn
At the 2025 SAFe Summit, moderator Dr. Steve Mayner opened a panel discussion with a number that landed like a punch:
"Ninety-five percent of organizations are realizing no measurable ROI from generative AI."
The room went quiet.
Not because the statistic was shocking—most people in enterprise AI have felt this reality in their bones. But because someone finally said it out loud, on a main stage, without the usual corporate hedging.
What followed wasn't a typical conference panel. It was a masterclass in why transformation fails—and what actually works when you stop treating AI adoption as a technology problem.
The panelists—Alison McCauley (author of Think with AI), Mik Kersten (CTO at Planview), and Laks Srinivasan (CEO of Return on AI Institute)—spent forty minutes dismantling the assumptions that keep organizations stuck. Their central argument?
The problem isn't the models. It's us.
More specifically: it's how we learn, how we organize, and whether we're brave enough to redesign work around human agency instead of just efficiency.
If you're a team leader caught between executives demanding AI results and team members resisting change, this conversation offers something rare: a map through the actual terrain.
The Pattern We Keep Repeating
Alison McCauley has been in Silicon Valley for 30 years, focused on emerging tech and how humans adapt to it. She's watched this movie before.
"When PCs were first introduced," she said, "we struggled to get ROI out of PCs."
The hardware worked fine. The software ran. But organizations didn't know how to use it. They treated adoption as a technology problem—install the machines, train people on the features, wait for productivity.
It didn't work.
What finally unlocked value wasn't better computers. It was humans learning to work differently: new workflows, new roles, new mental models for what work could be.
"It's not the raw technology horsepower that's going to make a difference," McCauley said. "It's us learning how to use it."
Mik Kersten reinforced the point with economic history. He drew a parallel to the "productivity paradox" of the 1980s and 90s—when IT spending soared but measurable output gains lagged by nearly two decades.
The technology was there. The learning curve wasn't.
"Individuals need to learn fast," Kersten said. "Organizations need to learn really fast—or they'll get left behind."
Here's the uncomfortable truth for middle leaders: Your executives think AI adoption is a deployment problem. Your team thinks it's a tool problem. Both are wrong.
It's a learning velocity problem.
And learning velocity doesn't come from more training modules or better prompts. It comes from culture.
Why "Data First" Kills AI Projects
Laks Srinivasan runs an institute that studies AI implementation across Fortune 100 companies. His research confirms what McCauley and Kersten intuited: the vast majority of failed AI projects start in the wrong place.
"Ninety percent of projects that fail," he said, "I bet you the first step is data."
Someone gets excited about the company's data assets. A leader says, "We should be able to do something with this." A team spins up, hires data scientists, builds models.
Then nothing. A bridge to nowhere.
Srinivasan's counter-framework is simple but radical: Outcome first, AI next.
Start with the human. Who's the "job executor"? What job are they trying to get done? What does success look like for them?
Only after you've defined outcomes in human terms should you ask whether AI is the right tool—and only then should you look at data.
He shared a case study: a utility company trying to use AI for power restoration after storms. First attempt failed because they started with available data and tried to find problems to solve.
Second attempt succeeded because they reframed around outcomes: minimize restoration time, reduce logistical costs, improve crew safety.
Same company. Same data. Different starting point.
"You have to meet them where they are," Srinivasan said. "There are some afraid they're gonna lose their jobs. You got to address that. There are some who can't wait to get their hands on this—you got to enable that."
Translation for team leaders: Stop asking "What can AI do with our data?" Start asking "What do our people need to accomplish—and where are they stuck?"
The second question opens up possibility. The first question opens up PowerPoints.
The Culture That Makes AI Work
McCauley has a concept she calls activating your learning network. It's the antidote to the top-down, training-first approach that most organizations default to.
"Our success depends on the strength of our learning network," she said. "We can no longer keep up as a person or even as a team."
Here's what that means in practice:
1. Find Your Passionistas
Before you know who your formal change agents are, identify the people already experimenting—even in their personal lives.
"People who are passionate about this technology hold such valuable information," McCauley said. "Find them, map them across the organization, connect them, develop dialogue with them."
These aren't necessarily your senior engineers or data scientists. They're the people showing up to meetings with AI-generated insights, or quietly using Claude to rewrite stakeholder communications, or experimenting with Midjourney for internal presentations.
They're distributed across functions. They're learning in the margins.
Your job isn't to control them. It's to activate them—and create channels for their learning to spread.
2. Create Forums for Experimentation
McCauley emphasized the need for structured spaces where people can play, fail, and share.
"Do no-code hackathons. Do science fair-like projects."
She mentioned a tech company that runs hackathons where participants code by voice only—hands literally tied behind their backs—to force creative problem-solving with AI.
The format matters less than the permission structure: We expect you to experiment. We want to hear what broke. We're learning together.
This is where most enterprise culture breaks down. Experimentation is tolerated in theory but punished in practice through performance reviews, budget cycles, and meeting culture that rewards certainty over discovery.
3. Amplify Through Storytelling
"Model the journey," McCauley said. "Talk about the bumps along the way."
Not just "Here's what worked"—but "Here's what I tried, why it failed, how I debugged it, and what I learned."
This is how organizational learning actually spreads: through narrative, not documentation.
For team leaders, this is your unlock. You don't need executive buy-in to start telling stories. You don't need a budget to activate passionistas. You don't need permission to create a weekly "show and tell" where people share experiments.
You need to model the behavior you want to see—and create the cultural container for others to do the same.
When Efficiency Threatens Humanity
Midway through the panel, Mayner posed a scenario designed to surface tension:
Imagine a company deploys AI to reduce agency spend on content creation by 30%. Financial goal: met. But the creative team feels their jobs have become rote. Their ingenuity is being designed out. They're miserable.
Is this a successful implementation?
The panelists' answers revealed a shared conviction: AI adoption without human agency is just automation theater.
McCauley reframed the premise entirely: "This can free human capacity. We will face choices about what to do with it."
She shared an example of a team that automated a weeks-long custom quote process down to hours. The team responsible? Happier—because they could now "allocate that time to strategic relationship building with their customers."
The pattern: AI eliminates the rote. What you do with the freed capacity determines whether transformation succeeds.
Kersten went further, proposing a new KPI that most enterprises would find radical: the happiness of the humans on each value stream.
"If you don't have happy people who feel empowered and autonomous and doing their work with some sense of purpose and mastery," he said, "you're not gonna have great people doing great work amplified by agents. You're gonna have mediocrity."
Laks Srinivasan connected it back to outcomes: "If you define the job executor as the person who's now bored, you didn't define it right. You didn't follow the right approach."
The synthesis: AI initiatives fail when they optimize for output (features shipped, costs reduced, tasks automated) instead of outcomes (problems solved, relationships deepened, capabilities unlocked).
And outcomes only matter when they're defined in human terms—what gets better for the person doing the work.
The Leadership Literacy Gap
Near the end of the discussion, McCauley said something that should make every executive uncomfortable:
"Leaders are not often using AI to truly advance their work—to do really strategic work. If you have not figured out how to use AI to bring meaning into your work, you need to begin. You cannot set a bold vision for AI unless you've experienced that."
This is the meta-problem.
Leaders commission AI strategies without personally experimenting. They delegate transformation to "digital innovation teams" while their own workflows remain unchanged.
Then they wonder why adoption stalls.
McCauley's diagnosis: "Until leaders start experimenting personally, they can't credibly lead transformation."
Not because they need to become prompt engineers. But because transformation is fundamentally an imagination problem—and you can't imagine new possibilities if you haven't experienced them yourself.
Srinivasan reinforced the accountability with an edge: "We know enough now. It's really on you if it fails. You can't blame the technology being early stage. There's enough bodies on either side of the road and enough research to study."
Translation: The learning materials exist. The case studies exist. The tools exist. If you're still waiting for someone else to figure it out, you've already lost time you won't get back.
What This Means for You
If you're a team leader navigating the gap between executive expectations and team resistance, here's what the panel offers:
Start with experimentation, not permission
You don't need a formal AI strategy to activate your passionistas. You don't need budget approval to start telling stories about what's working.
Create space. Model curiosity. Share your own failures.
Reframe outcomes in human terms
Stop asking "How can AI reduce costs?" Start asking "What do our people need to accomplish—and where are they stuck?"
The first question leads to automation theater. The second leads to capability building.
Measure happiness alongside productivity
If people feel disempowered by your AI implementation, you're not transforming—you're just optimizing toward mediocrity.
Kersten's happiness metric isn't soft. It's strategic. Empowered, autonomous people amplified by AI create exponential value. Demoralized people with AI access create compliance and churn.
Lead by learning in public
The fastest way to shift culture is to model the behavior you want to see.
Share what you're experimenting with. Talk about what broke. Ask for help. Create permission through example.
The Real Transformation
The panel closed with three final thoughts—one from each panelist:
Kersten: "Shift from output to outcome."
Srinivasan: "Outcome first, AI next."
McCauley: "Open your mind to what's possible with this."
Taken together, they describe a transformation that's less about technology and more about organizational learning velocity.
The companies that will thrive in the age of AI aren't the ones with the biggest models or the most data.
They're the ones that learned how to learn faster than their competitors.
And learning velocity doesn't come from top-down mandates or training programs. It comes from culture: the stories people tell, the experiments they're allowed to run, the permission structure that says "we expect you to try things and share what you discover."
You don't need executive buy-in to start building that culture.
You just need to begin.
Watch the Full Discussion
The forty-minute panel discussion between Alison McCauley, Mik Kersten, Laks Srinivasan, and moderator Dr. Steve Mayner covers far more ground than we can capture here—including specific frameworks for outcome-first thinking, case studies of what works (and what fails), and tactical advice for leaders at every level.
Watch the complete session below:
The next time someone asks why AI adoption is failing in your organization, don't point to the tools. Point to the learning environment.
If people aren't experimenting, it's not because they don't understand the technology.
It's because the culture hasn't given them permission to learn.
And permission, as it turns out, is the one thing you can grant without anyone's approval.
2
Comments (0)
Popular
