The challenge of moving AI from prototype to production [Q&A]

More organizations are turning to AI to assist in their digital transformation efforts, but many projects get stuck in the pilot phase.

That’s not necessarily a sign of failure though. Rather, it reflects that AI is still in its formative stage, with its most transformative impact still ahead. We spoke to Nadav Eiron, SVP of cloud engineering at Crusoe to learn more about how AI can transition from experimentation to integration, and from potential to permanence.

BN: How do you see AI initiatives progressing today?

NE: And we’re already seeing what’s possible when companies make the leap into production systems. Accenture’s 2024 research shows that organizations leading with AI -- where at least 30 percent of processes are AI-enabled -- achieve 50 percent higher productivity and 25 percent greater profitability than their peers.

In the end, progress won’t hinge on building bigger models -- it will depend on broader access. The real breakthrough will come when the people closest to the work can shape how AI is applied. When AI becomes a tool for many, not just the technically fluent few, its transformation from promise to impact will truly begin.

BN: For any AI initiative bottlenecks, what are the causes?

NE: A big part of the problem isn’t technical at all -- it’s organizational. In most companies, the people who understand the daily friction points best aren’t the ones empowered to shape how AI is applied. The logistics planner. The warehouse manager. The plant technician. They ‘live’ the problems. Yet the tools being built often don’t reflect that lived experience.

Not because no one’s asking. But because the systems we’ve built still require too much technical fluency. Talk to someone in operations, and you’ll often hear a version of the same story: “I know what needs fixing. But I can’t explain it in a way that gets into the system.”

And that’s a missed opportunity. The people closest to the work often have the clearest view of where AI can drive real change. Imagine the impact if those insights could directly shape the tools -- if frontline experts could not only use AI, but help build it. That’s when AI shifts from being applied to the work, to being shaped by it.

True, platforms such as AWS Bedrock are simplifying access -- but they still assume a user base knows how to code or to fine tune a model. To a domain specialist, it can feel like stepping into someone else’s workshop. There are tools everywhere, but no instruction manual in sight.

BN: And the opposite, what has AI been missing improves its success rate ?

NE: Major leaps in computing haven’t come just from faster chips or better algorithms -- they’ve come from abstraction. Think about how programming evolved: from cryptic assembly code to high-level languages, from monoliths to modular microservices. Each shift made development faster, more scalable, and more inclusive, which in turn brought about the next wave of innovation..
AI hasn’t had that moment yet.

Right now, building something useful with AI means wrangling a complex stack: foundation models, vector databases, prompt engineering, evaluation frameworks. It’s duct tape and trial-and-error. Even getting a basic proof of concept into shape can take weeks, not because the idea is bad, but because the plumbing is a mess.

We don’t need to dumb things down. But we do need shared patterns. Interfaces that don’t hide the machine, but shield users from its sharpest edges. Good abstractions let people work at the level of intent. That’s what AI is missing.

BN: How can organizations scale their AI initiative in a strategic manner? Any recommendations?

NE: The pursuit of ever-larger models continues -- driven by benchmark competitions, academic incentives, and the gravitational pull of headline-grabbing breakthroughs. But for most enterprises, that kind of progress is starting to feel increasingly detached from day-to-day reality.

The real leap forward won’t come from scaling up models. It will come from scaling their use. Many of today’s models are already more than capable of solving meaningful business problems. What’s missing is deployment, contextualization, and integration. The gains that matter now aren’t in model architecture, but in operational relevance: connecting AI to the processes, data, and people that actually drive the business.

Yet too often, organizations chase the next frontier in capability instead of extracting value from what’s already within reach. In the process, practical use cases—especially in sectors like manufacturing, logistics, or the public sector -- get overlooked, not because they lack potential, but because they don’t align with what the market currently hypes up.

A more balanced approach would focus on right-sized AI -- smaller models, built around real-world tasks, that don’t require a data scientist to configure.

BN: Historically speaking, can you compare the AI explosion to other past enterprise tech trends?

NE: This isn’t the first time a transformative technology has outpaced its foundations. In the early web era, the vision was there–online shopping, digital media, video–but the infrastructure wasn’t. The web didn’t scale until we had standards, protocols, middleware. They enabled broad adoption, made it usable beyond specialists, and embedded it into everyday workflows -- from office communication to supply chains. That’s what real transformation looks like: not just potential, but practical integration at scale.

AI is now at a similar point with AI. The demos are dazzling. But the systems to make them reliable, explainable, and interoperable are still under construction.
The danger? A widening gulf between what’s possible in theory and what’s practical in the field. Without better scaffolding, the promise of AI could stall in a swirl of pilot projects and PowerPoints.

If there’s a single insight that should shape the next wave of AI development, it’s this: progress won’t come just from better models. It will come from more people being able to do more with the ones we already have and shape the ones to come.

Image credit: Tongsupatman/Dreamstime.com