At Sierra Ventures, conversations with enterprise technology leaders are one way we stay close to how AI is actually being deployed inside large organizations, beyond the demos and headlines.

Mark Fernandes recently sat down with Earl Newsome, Chief Information Officer at Cummins, a global power technology manufacturer with over $30 billion in revenues, to discuss where AI is delivering real value inside industrial enterprises, how to set the right expectations, and what it takes to build a durable competitive advantage in a world that demands 90-day decisions.

It's Not a Failure Rate, It's a Learning Rate

One of the sharpest reframes in the conversation came early. Earl pushed back directly on the studies, citing high failure rates for enterprise AI initiatives, arguing they mischaracterize what's actually happening.

"It's not a high failure rate, it's a high learning rate."

At Cummins, the approach has been to let business units and functions experiment broadly, then send a working group out to harvest what works and build on it. The areas where Cummins is focused on extracting outsized value: supply chain, shared services, co-engineering and product development, and sales enablement and customer engagement.

The underlying model Earl described is simple: learn once, learn forward, learn cheaply, learn fast. If one in ten experiments generates a significant value, that's a good ratio. The other nine teach you how to move faster on the next round.

The Hardest Part Is the Pace of Change

When asked what's been harder than expected, Earl didn't point to technology or data. He pointed to speed itself.

"In the mainframe days, we made 20-year decisions. In the client-server days, we made 10-year decisions. In the internet-and-cloud days, we made three- to five-year decisions. In this AI world, we need to make 90-day decisions."

That shift has implications across the entire enterprise operating model: how you procure, contract, architect, and implement. Solutions need to be modular and composable so that a 90-day pivot doesn't create downstream damage. It's a fundamentally different way of building and buying technology.

This pattern shows up consistently across Sierra Ventures' enterprise network. The organizations moving quickly on AI are the ones that invested early in flexible architecture, not just models.

The Competitive Moat Is Composability

Earl's view on durable competitive advantage was direct: whoever figures out AI in their industry wins their industry. But winning isn't just about adopting AI first. It's about building the capability to version rapidly, connect systems, and operate in what he described as an inevitable multi-agent, multi-model future.

"You need to build solutions that enable you to live in a multi-agent, multi-model world. It needs to be disciplined, it needs to be observable, but it needs to be connectable and composable."

That composability, built on a bedrock of high-quality data and clean architecture, is what separates point solutions from platforms. It's also where Earl sees the clearest opportunity for startups building into the enterprise: solutions that are rentable, versionable, and plug-and-play, with a versioning model that improves continuously rather than requiring a rip-and-replace.

What Founders Should Pay Attention To

Earl's advice for startups was grounded in the operational reality of large enterprises moving at 90-day cycles. The ask isn't just a great product. It's a great product that delivers impact quickly, improves continuously, and fits into an ecosystem rather than requiring one.

"Every day or every other day, there's an announcement, there's new models coming out, there's agents that are being built. This is the slowest it'll ever be."

The founders who will win enterprise deals in this environment are those who understand their customers aren't evaluating a static solution. They're evaluating a trajectory.

Watch the video: