'Actual Engineering Is Getting Harder To Fake': Why Liberty Mutual Keeps Engineering Rigor At The Core Of Its AI Strategy

AI’s impact on engineering goes far beyond automated code generation.

For over twenty years, software engineering operated on a predictable rhythm. The shift to the public cloud and the adoption of CI/CD pipelines may have changed underlying infrastructure and accelerated how fast new technologies were introduced to the business. But teams still built and deployed tools for end users they may never have communicated with.

Now, as AI lowers the technical barrier to creating custom software, more companies are shifting development in-house. And engineering teams are increasingly expected to partner with domain experts from across the business to scale tailored solutions.

“AI is disrupting the technology changing the way we operate and is reshaping our culture,” said Tony Marron, Global Head of Engineering Capability at Liberty Mutual Insurance. “Whenever you have three significant changes happening at once, it can be disorienting.”

That shift is also changing how engineers work with colleagues across the business. “The old model was building a solution for someone. Increasingly, we’re building it with them,” Tony said.

While AI gives enterprises more control over their software arsenal, the technology also quickly exposes the gaps in existing infrastructure environments. As more employees partner with each other and AI agents to create software, an organization’s engineering culture and foundation become more important than ever.

“AI is definitely making it easier to build solutions. But strong engineering is getting harder to fake,” said Tony. “AI is not removing the fundamentals; it’s exposing them. The quicker you move, the quicker you’ll spot weak engineering and weak architecture.”

CIOnews spoke with Tony to learn how Liberty Mutual scales AI adoption across the enterprise while keeping engineering rigor at the center of its tech strategy.

The Three Levels of AI Adoption

At Liberty Mutual, AI adoption is broken into three levels: individuals, teams, and the organization as a whole. The approach is designed to meet employees where they are while building AI capabilities so teams can “learn together and move in the same direction,” said Tony.

• Everyday AI (Individual): Employees are given tools and space to experiment, build their AI understanding and get more comfortable applying the technology in their work.
• Supercharge (Team): AI is embedded within workflows, systems, and platforms, crossing multiple teams in the process.
• Rethink (Organizational): Processes are completely reshaped around AI using agentic engineering, with humans and agents working together across workflows.

Each level operates within appropriate security, governance and responsible AI controls, with human oversight and accountability remaining central to the approach.

Underpinning all three levels is a common, composable architecture that gives Liberty Mutual the ability to design workloads based on evolving criteria across reasoning power, cost, and speed.

“If you lock yourself into one relationship or partnership, it can be very difficult to take advantage of the different dimensions and capabilities that are evolving all the time,” said Tony. “Our approach allows us to bring our employees along on the journey, and that's the most important thing. And it allows us to mature our skills and architecture together and be really deliberate about both.”

Strategic Evaluation and Agentic Engineering

When evaluating new technologies, the IT team starts with the business problem rather than the technology itself, working with employees, partners, and other stakeholders to understand the outcome they’re trying to improve and where technology can add the most value. Candidates are tested, and the most promising options move to formal experimentation where solutions run alongside existing technologies to evaluate ecosystem compatibility. As the cost and time required to experiment falls, Tony said the discipline around what ultimately scales becomes even more important: “Experiment freely. Scale deliberately.”

For example, when Tony’s team explored potential agentic engineering platforms, they found several didn’t provide the level of composability they were looking for. They needed the flexibility to pull in different frameworks and components.

“We implemented an agentic engineering factory that changes not only an engineer’s day-to-day work and responsibilities, but how we build solutions and partner with colleagues across different functions,” Tony said.

As AI takes on more of the mechanics of software development, engineers can spend more time framing problems, designing systems, evaluating outputs and making decisions about what ultimately moves into production. For Tony, that makes engineering judgment and deep technical expertise more important, not less.

Speed Is Built on Unseen Foundations

Like many organizations, as Liberty Mutual expands its use of AI, it is evolving the traditional build-versus-buy decision. Increasingly, Tony sees the answer as a combination of building, buying and composing capabilities rather than choosing one approach over another. Because the technology lowers the barriers to software creation, new innovations can launch in weeks or months rather than years. For example, Liberty Mutual recently built and deployed a new conversational AI auto insurance quoting experience in roughly three months.

“The reason we were able to do that quickly is because of the work the teams had done over the years to make sure our services were API-first, ” Tony said. “We had done the foundational work, so when we decided to apply these new AI capabilities, we could move quickly.”

Along with owning the product roadmap, building with new AI tools helps teams better understand the technology, enabling them to make smarter buying decisions down the road. Ultimately, for Tony, the leadership challenge boils down to a core question: What do we need to control, and what authority or judgment are we comfortable delegating?

“It’s less about buy vs. build—you have to do both,” said Tony. “You can buy the policy engine, but you have to own the rules and policies it enforces.”

AI is only as good as the engineering foundation beneath it. Fragile connections, siloed systems, and rigid architectures will inevitably undercut the speed and independence that AI promises. But beyond faster code generation, the real promise of enterprise AI lies in dismantling the traditional walls between technical teams and business domain experts.

By pairing a composable, API-first architecture with a culture of shared experimentation, engineering teams can move rapidly without surrendering control or governance. The tools will continue to evolve, but the ability to ground complex technology in sound architecture to drive real business outcomes remains the true mark of strong engineering.

‍