solving the AI paradox: why “defensive modernization” is no longer enough for consumer and retail leaders.

Over the past year, I’ve had a version of the same conversation with CIOs across some of the largest retailers and CPG organizations in the world. The consensus: AI is live. Use cases are in production. On paper, the investment is showing up in specific pockets of efficiency. But they simply aren’t seeing a different company in the mirror. 

This is the “AI paradox,” a state where increased investment and efficiency gains don’t translate to any real change in organizational DNA. And it persists largely because many leaders lack the frameworks to measure true productivity gains or the visibility into how deeply AI tools are woven into their workforce’s daily habits.

The result is a fork in the road. On the one hand, organizations are using AI for “defensive modernization”: optimizing the model they already have for better forecasting, faster cycle times, incremental efficiency. On the other is a smaller group doing something fundamentally different. They’re rethinking how execution is structured across systems, teams and partners. Rebuilding, not just accelerating, whole workflows to solve the paradox and go from speed to structural transformation. 

Both groups are investing. Both are moving. But they’re not going to land in the same place.

the trap of visible progress. 

Security and compliance are driving a large part of the first group’s defensive stance. As machine-led threats become more sophisticated, Zero Trust and AI-enabled security platforms have moved from long-term priorities to immediate requirements. Then, there’s the pressure to show visible progress with AI. Copilots, assistants and targeted automation are being deployed quickly, often in areas where value is easy to demonstrate and risk is contained. 

All of this is necessary. In many cases, overdue. But taken together, they tend to reinforce the status quo. We see this in the value gap: 80% of companies use GenAI; only about 40% report EBIT impact from AI.¹ That gap doesn’t come down to effort but to where that effort is being directed. What happens when front-end gains outpace improvements to data, systems and process design? Progress becomes visible but doesn’t compound. Initiatives scale slowly, if at all. You can make a process 10% faster, but if the underlying structure is rigid, the business hasn’t fundamentally moved.

designing for evolution: the intelligent core. 

The organizations pulling ahead may be spending more for further savings and optimization in the future. But more importantly, they’re spending more deliberately because they realize that modern AI can’t thrive on legacy ERP and other environments that weren’t built for flexibility, real-time decisioning or rapid integration. And instead of working around this challenge, their focus is on building an intelligent core: one designed to support continuous change rather than resist it.

These organizations are even using AI to help with the work of building the infrastructure it needs to run on. Refactoring legacy code. Mapping data across systems. Automating upgrade cycles and identifying where redundancy actually exists across tech estates.

But infrastructure is only half the battle. The organizations leading this era also design for evolution across people and technology simultaneously. If a role looks the same today as it did 18 months ago, it’s already falling behind. As agentic systems take on routine execution, human value is moving upstream toward judgment, oversight and the design of how systems and people work together.

closing the implementation gap. 

Tools like Cursor and Claude Code have compressed the distance between idea and execution. Speed is becoming a commodity. What isn’t broadly available is the organizational layer that makes these AI-driven gains compound. Without clear architectural governance and ownership standards that travel as fast as the builds do, faster development just produces faster fragmentation. More solutions, less coherence, and a new layer of technical debt accumulating underneath the progress. The bottleneck? The implementation gap. 

Bridging this gap means moving away from a “software-first” or “headcount” mentality toward an AI-talent-first architecture. This is the core of AI talent-as-a-service (ATaaS), where true value isn’t in the software but the architects, prompt engineers and strategists who integrate it into a business’s DNA. That's where the real leverage is, and it’s why workforce planning and AI strategy can no longer be separate conversations.

the leap to agentic systems.

Beyond task-level automation, leaders in this space are investing in agentic systems. Unlike traditional robotic process automation, these systems aren’t limited to predefined rules. They operate in variable environments, making decisions within defined guardrails and coordinating across systems in real time.

The impact shows up quickly in operations-heavy functions. In merchandising, virtual analysts handle routine monitoring, freeing humans to focus on growth. In the supply chain, logistics agents adjust routing based on real-time conditions. This is where the structural shift becomes visible: stores repositioned as fulfillment nodes and AI coding agents enabling customized, AI-native applications that reduce reliance on rigid platforms. 

As the cost of making changes comes down, the pace of change increases. Systems that once felt fixed start to look flexible.

conclusion.

The next phase of AI adoption won’t be defined by who deploys the most tools but by whoever redesigns their organization most effectively. Success can no longer be measured in silos. It must be measured by how the entire company has evolved its ability to respond to change.

Most organizations know which of two groups they’re in. However, moving from defensive modernization to structural advantage requires a deliberate path:

• The AI audit: Identifying where AI can drive a structural shift rather than just an incremental gain.
• Redefining the work: Evolving roles and workflows so human judgment and AI execution work in a compound loop.
• Rapid execution: Deploying specialized talent like ATaaS squads to build, test and scale solutions in weeks, not years.