Caterpillar’s Digital Chief Ogi Redzic on How A.I. Is Redefining Construction Work

Beneath attention-grabbing headlines about driverless taxis and laundry-folding robots, some of the most consequential deployments of physical A.I. are actually unfolding quietly in construction and earthmoving. One company leading this shift is Caterpillar.

In 2018, when former CEO Jim Umpleby set out to transform the traditional heavy-equipment manufacturer into a modern, service-driven enterprise, he brought in Ogi Redzic, a seasoned automotive executive then leading connectivity and autonomy at Renault-Nissan, to unify Caterpillar’s scattered digital efforts. Redzic spearheaded the development of Helios, a central platform that now manages connected machines worldwide. That early work proved prescient and laid a strong data foundation just in time for the A.I. wave.

Today, Redzic leads Cat Digital, a 3,000-person division building the digital tools behind Caterpillar’s equipment. His work not only determines how the world’s largest heavy equipment maker uses A.I., but shapes how the technology will redefine construction work itself. As the industry faces acute labor shortages, with nearly 40 percent of current operators set to retire by 2031, Redzic sees autonomous tools as a vital way to bridge this gap without displacing existing jobs. “Humans won’t be out of the loop anytime soon,” he notes.

Redzic spoke with Observer about how Cat Digital is deploying A.I. to assist operators on the ground, why clean data matters, and how A.I. is changing the way his own team builds software.

The following conversation has been edited for length and clarity.

Observer: You run a digital unit with over 3,000 employees at a company most associated with heavy machinery. What does your team do, and how does that work show up in the machines and on a construction site?

Ogi Redzic: My team builds off-board technology—everything we offer to customers and dealers. There’s technology running primarily on the cloud: a collection of apps we build to support those machines. When my team develops something, it’s almost always to make it easier for our customers to operate, to make job sites more productive, or to make equipment easier to maintain.

For example, we make software for equipment management. If you’re a fleet operator on a job site with 20 or 30 pieces of Cat equipment, we have an application to see where those assets are, how they are working, which ones need maintenance, and which ones have urgent issues.

We also make it very easy for customers to buy parts or schedule service. We put serialized QR codes on all our assets so you can scan one, search for an air filter, and get the exact part that works on your machine. Beyond e-commerce, if we notice a need for a transmission repair on an asset, we can organize that digitally with your local dealer so you can schedule service with a single click.

Caterpillar equipment lasts a very long time, so we need to provide support for equipment built in the ’60s, ’70s and ’80s. We help field technicians with knowledge of all previous generations of machinery. They can pull up a digital app that says, “To replace the hydraulic pump on this asset built in 1979, here are the steps.”

As we get new capabilities, we can do more of that using A.I. glasses and other mediums.

We often see headlines about self-driving cars or humanoid robots, but some of the most impactful A.I. applications are happening in agriculture and construction. How is A.I. changing Caterpillar’s equipment and machinery?

Machines are going to become increasingly intelligent and be able to operate on simple commands. We are already applying solutions like the “Cat AI Assistant” to make it easier for operators to interact with our machines: a less experienced operator can use a voice interface for real-time coaching and training without needing a senior operator sitting next to them.

We have remotely operated machines that will increasingly run on A.I. to become fully autonomous. We’ve done this in mining for decades, with hundreds of millions of kilometers driven autonomously without safety incidents. Construction sites are more complex than mining sites, so our first focus is “task autonomy”—allowing individual machines to execute specific assigned tasks.

In five to ten years, you will likely see A.I. coordinating multiple machines on a construction site. On a typical construction site, up to 40 percent of the time is unproductive—people waiting for another machine to finish, searching for original design files, or experiencing idle delays. An A.I. layer can orchestrate site operations, measure progress and adjust workflows to maximize productivity.

Is this going to make a construction worker’s job easier or more demanding?

It will be a transition. Right now, there is a severe operator and technician shortage. Construction jobs take twice as long to fill as they did before COVID. About 40,000 technicians retire in the U.S. every year, and nearly 40 percent of current operators will retire by 2031.

Automation helps fill that gap. A.I. will coach newer operators to get them up to speed faster while automating repetitive tasks, so you don’t need as many people to complete a job. Workers will transition from manually operating machines to overseeing autonomous workflows as site supervisors. Humans won’t be out of the loop anytime soon.

You joined Caterpillar in 2018, brought on by former CEO Jim Umpleby. At the time, Caterpillar had digital efforts scattered across different teams, and you were tasked with bringing everything under one umbrella. It was years before generative A.I. became mainstream. What drove the urgency to unify those efforts?

One of Jim’s most important strategic initiatives was turning Caterpillar from a traditionally OEM-focused company—shipping new equipment—to a more service-focused one. We placed a strong emphasis on supporting our customers beyond the point of purchase to help them become more efficient and productive.

As a large global enterprise, we have three different segments and serve many industries. Historically, each business unit created digital products to satisfy its own needs. Each had its own view of the customer, its own data set and its own applications. That works up to a point, but if you want a unified view of the customer across multiple dealers and departments, you need a common set of services. Otherwise, digital solutions won’t feel like they’re coming from the same company.

How did you achieve that, and how did that early work prepare you for the A.I. boom?

Over the first couple of years, we were focused on structuring our data, because you can’t build good digital products without a strong foundation. We built a platform called Helios, which allowed us to cleanse and organize the data. Helios now stores about 18 petabytes of data and processes roughly 700 billion data pipeline transactions every month. On top of Helios, we built individual apps for e-commerce, condition monitoring to catch equipment issues early, and inspections.

If you had asked me in 2018 where we would be today, I don’t think anyone would have predicted this level of A.I. adoption. But even back in 2018, we built our data platform with traditional machine learning in mind. A lot of our condition-monitoring analytics rely on machine learning to identify patterns and predict when a failure might occur within a certain timeframe.

We didn’t expect the sudden advancements in generative A.I. or agentic A.I., but because our data was so well-organized, we were able to jump on the technology as soon as it became available. Our A.I. system was brought to market in less than a year from when we started.

How has A.I. changed the software side of Caterpillar’s products? How does your team work with A.I.?

We are now putting platform services into production that were entirely developed by A.I., from concepting and spec generation to design, coding, testing and deployment. We can now accomplish three times the work with only 20 percent to 30 percent more people.

This forces us to rethink our agile development processes. Instead of spending months writing rigid specs before coding, you can prototype and test ideas almost instantly. Writing code is becoming less critical than knowing how to guide, debug and govern A.I.-generated systems.

What does that mean for hiring? Are you hiring less, or using different criteria when screening candidates?

We haven’t stopped hiring, but we are looking for “A.I.-native” talent, or people who instinctively know how to build and think within an A.I.-driven context. We recruit heavily from top computer science programs for internships and early-career roles, and these new hires hit the ground running immediately.