If you think deploying AI is just about moving fast and breaking things, you haven’t earned the right to use the tech yet. In a heavily regulated world, the role of a leader has split into two high-intensity lanes: Offense and Defense. Monica Caldas, Global CIO of Liberty Mutual, has turned this framework into a blueprint for 2026. You cannot play a winning offensive game building new features and scaling AI if your defensive line is leaking data or running on unstable systems.
Earning the Right to Innovate
Modernization is not a “lift and shift” project; it is a full-scale transformation. If your structured and unstructured data are still trapped in isolated silos, you will hit a wall with Generative AI almost immediately. You have to play defense by securing and stabilizing your foundations first. Only then do you earn the right to play offense. Without a secure, stable system, your AI ambitions are just a liability waiting to happen.
The ‘Jobol’ Trap: Why Architecture Matters More Than Code
Generative AI is not a magic wand. If you try to use it to simply turn old COBOL code into Java, you end up with “Jobol,” a messy, Frankenstein-like architecture that lacks the security protocols and modern functionality required for production. AI can spit out code, but it cannot build a modern architecture for you. To win in 2026, you need to clean up and retire the old systems rather than just painting over them with new AI-generated scripts.
Deploying Internal Agents (The ‘Libby’ Model)
At Liberty Mutual, the offensive strategy started internally with an AI agent called Libby. By attaching this agent to a massive knowledge database and environment instrumentation, the company moved away from manual help-desk workflows.
- The Result: Routine issues are predicted and solved automatically.
- The Talent Shift: Help desk employees aren’t being replaced; they are being redeployed to tackle the high-value backlogs that actually require human intuition.
The Seniority Surge in Software Development
AI is currently supporting roughly 35% of the software development life cycle, but it isn’t a shortcut for everyone. Senior engineers are using AI to “fly,” significantly increasing their output and quality. However, junior engineers still require heavy mentoring to prevent them from relying blindly on AI-generated output. Productivity in 2026 is multidimensional; it is about better decision-making and higher-quality products, not just “more” code.
The Bottom Line
Stop treating AI as a standalone miracle. Success requires a Responsible AI Steering Committee to navigate the risks and a workforce trained to spot hallucinations before they hit production. If your offense is faster than your defense, you are just building a faster way to fail.
