When AI Fails, It's Not the Code — It's the Culture

There's a moment that happens in almost every organization rolling out a new AI tool. The system goes live. The dashboards light up. The metrics look promising. Then, three months later, someone in senior leadership asks the question nobody wants to hear: "Why isn't anyone actually using this?" I've watched this play out again and again, and it wasn't until my conversation with Professor Carmen Pires Migueles that I finally understood what was really happening. The technology wasn't failing. The culture was rejecting it like a bad transplant.

Professor Migueles has spent decades studying what actually moves organizations forward, not from the comfort of theory but from the trenches of companies like Petrobras, Transpetro, and Citibank, and even as Municipal Secretary of Culture in Duque de Caxias. She holds a PhD in Sociology of Organizations from Sophia University in Tokyo and completed postdoctoral work examining the invisible forces that shape how people work. When she told me, "We like to talk about AI as if it's some unstoppable tidal wave, but the real lever of change isn't the technology — it's the culture it enters," I felt something click. We've been asking the wrong question. We keep asking what AI can do. We should be asking what kind of organization can actually absorb it.

The evidence is everywhere once you start looking. I've seen companies spend millions on predictive analytics platforms that sit unused because middle managers fear the transparency will expose their decision-making. I've watched knowledge management systems fail not because the search algorithms were weak, but because employees had spent years hoarding information as currency. The technology worked perfectly. The culture suffocated it. What Professor Migueles helped me see is that AI doesn't enter a vacuum. It enters a living ecosystem of norms, power structures, rituals, and unspoken rules. If that ecosystem is rigid, hierarchical, or built on distrust, even the most elegant algorithm will be neutralized. The system may run, but it won't transform anything because transformation isn't a technical process. It's a social one.

This explains why some organizations thrive with AI while others stagger. It's not about who has the best models or the biggest budgets. It's about whether the culture rewards experimentation or punishes deviation. Whether it encourages people to surface problems or hide them. Whether it treats data as a shared resource or a weapon. Professor Migueles pointed out something I hadn't fully articulated: organizations are fundamentally about people making sense of their world together. When AI challenges how they've always made sense of things — how they've always known who matters, what counts as success, what feels safe — resistance isn't irrational. It's human. And if leadership doesn't address that human dimension, no amount of technical sophistication will matter. You can't install culture like you install software.

What does this mean for those of us leading AI initiatives? First, it means we need to stop pretending implementation is a technical project. It's a cultural intervention. Before you deploy a single model, ask: Does our culture reward transparency or opacity? Do people feel safe admitting what they don't know? Are failures treated as learning opportunities or career killers? If the answers reveal a brittle, defensive culture, your AI project is already in trouble. You're not just introducing a tool. You're asking people to change how they relate to information, authority, and each other. That requires psychological safety, trust, and a leadership team willing to model the vulnerability they're asking of everyone else.

Second, we have to design for the culture we have, not the one we wish we had. I used to think the right move was to build the ideal system and let it pull the culture forward. Professor Migueles helped me see that's backwards. If your culture is risk-averse, start with low-stakes AI applications where failure is cheap and learning is visible. If your culture is siloed, begin with cross-functional pilots that force collaboration but keep the scope contained. If your culture worships credentials and hierarchy, bring in respected internal champions who already have social capital. You're not dumbing down the technology. You're respecting the reality that adoption is a social process. People don't adopt tools because they're powerful. They adopt tools because trusted peers are using them, because they see relevance to their own work, because the change feels manageable rather than threatening.

Third, and perhaps most important, we need to recognize that culture isn't a fixed obstacle to overcome. It's the medium through which change happens. The most successful AI transformations I've witnessed didn't bulldoze existing culture. They worked with it, identifying the parts that could accelerate adoption — collaborative norms, curiosity about customers, pride in solving hard problems — and amplifying those. They also identified the parts that would block it — fear of automation, distrust between functions, incentive structures rewarding individual heroics over team outcomes — and addressed them directly, not with another memo but with real shifts in behavior, starting at the top. Culture isn't an enemy of innovation. It's the terrain. And the best leaders don't fight the terrain. They learn to read it.

I keep coming back to something Professor Migueles said that felt almost radical in its simplicity: organizations are systems, yes, but they're made of people. And people don't change because you give them better tools. They change because they're invited into a different story about who they are and what's possible. AI can be part of that story, but only if we're willing to do the harder, slower work of cultural transformation first. The code will run. The question is whether anyone will let it matter.

If your AI isn't working, don't debug the algorithm first. Debug the culture.# When AI Fails, It's Not the Code — It's the Culture

There's a moment that happens in almost every organization rolling out a new AI tool. The system goes live. The dashboards light up. The metrics look promising. Then, three months later, someone in senior leadership asks the question nobody wants to hear: "Why isn't anyone actually using this?" I've watched this play out again and again, and it wasn't until my conversation with Professor Carmen Pires Migueles that I finally understood what was really happening. The technology wasn't failing. The culture was rejecting it like a bad transplant.

Professor Migueles has spent decades studying what actually moves organizations forward, not from the comfort of theory but from the trenches of companies like Petrobras, Transpetro, and Citibank, and even as Municipal Secretary of Culture in Duque de Caxias. She holds a PhD in Sociology of Organizations from Sophia University in Tokyo and completed postdoctoral work examining the invisible forces that shape how people work. When she told me, "We like to talk about AI as if it's some unstoppable tidal wave, but the real lever of change isn't the technology — it's the culture it enters," I felt something click. We've been asking the wrong question. We keep asking what AI can do. We should be asking what kind of organization can actually absorb it.

The evidence is everywhere once you start looking. I've seen companies spend millions on predictive analytics platforms that sit unused because middle managers fear the transparency will expose their decision-making. I've watched knowledge management systems fail not because the search algorithms were weak, but because employees had spent years hoarding information as currency. The technology worked perfectly. The culture suffocated it. What Professor Migueles helped me see is that AI doesn't enter a vacuum. It enters a living ecosystem of norms, power structures, rituals, and unspoken rules. If that ecosystem is rigid, hierarchical, or built on distrust, even the most elegant algorithm will be neutralized. The system may run, but it won't transform anything because transformation isn't a technical process. It's a social one.

This explains why some organizations thrive with AI while others stagger. It's not about who has the best models or the biggest budgets. It's about whether the culture rewards experimentation or punishes deviation. Whether it encourages people to surface problems or hide them. Whether it treats data as a shared resource or a weapon. Professor Migueles pointed out something I hadn't fully articulated: organizations are fundamentally about people making sense of their world together. When AI challenges how they've always made sense of things — how they've always known who matters, what counts as success, what feels safe — resistance isn't irrational. It's human. And if leadership doesn't address that human dimension, no amount of technical sophistication will matter. You can't install culture like you install software.

What does this mean for those of us leading AI initiatives? First, it means we need to stop pretending implementation is a technical project. It's a cultural intervention. Before you deploy a single model, ask: Does our culture reward transparency or opacity? Do people feel safe admitting what they don't know? Are failures treated as learning opportunities or career killers? If the answers reveal a brittle, defensive culture, your AI project is already in trouble. You're not just introducing a tool. You're asking people to change how they relate to information, authority, and each other. That requires psychological safety, trust, and a leadership team willing to model the vulnerability they're asking of everyone else.

Second, we have to design for the culture we have, not the one we wish we had. I used to think the right move was to build the ideal system and let it pull the culture forward. Professor Migueles helped me see that's backwards. If your culture is risk-averse, start with low-stakes AI applications where failure is cheap and learning is visible. If your culture is siloed, begin with cross-functional pilots that force collaboration but keep the scope contained. If your culture worships credentials and hierarchy, bring in respected internal champions who already have social capital. You're not dumbing down the technology. You're respecting the reality that adoption is a social process. People don't adopt tools because they're powerful. They adopt tools because trusted peers are using them, because they see relevance to their own work, because the change feels manageable rather than threatening.

Third, and perhaps most important, we need to recognize that culture isn't a fixed obstacle to overcome. It's the medium through which change happens. The most successful AI transformations I've witnessed didn't bulldoze existing culture. They worked with it, identifying the parts that could accelerate adoption — collaborative norms, curiosity about customers, pride in solving hard problems — and amplifying those. They also identified the parts that would block it — fear of automation, distrust between functions, incentive structures rewarding individual heroics over team outcomes — and addressed them directly, not with another memo but with real shifts in behavior, starting at the top. Culture isn't an enemy of innovation. It's the terrain. And the best leaders don't fight the terrain. They learn to read it.

I keep coming back to something Professor Migueles said that felt almost radical in its simplicity: organizations are systems, yes, but they're made of people. And people don't change because you give them better tools. They change because they're invited into a different story about who they are and what's possible. AI can be part of that story, but only if we're willing to do the harder, slower work of cultural transformation first. The code will run. The question is whether anyone will let it matter.

If your AI isn't working, don't debug the algorithm first. Debug the culture.