Joe Rose, president of strategic technology provider JBS Dev, is pushing back against a widely held assumption about generative and agentic artificial intelligence systems. He argues that the belief data must be flawless before any AI workload can begin is a misconception.
According to Rose, vendors and consultants often recommend building massive data lakes or undertaking multiyear data transformation programs. Such advice can leave executives uncertain about how to proceed. The reality, he contends, is more pragmatic.
Modern Tools Handle Imperfect Data
Rose points out that today’s tooling is better equipped than ever to manage poor-quality data. He notes that large language models can interpret even half-written prompts with remarkable accuracy. This capability, he says, allows organizations to work with existing data rather than waiting for perfect datasets.
The unpredictable nature of AI models, however, requires safeguards. Rose emphasizes the importance of human oversight to catch and correct erroneous outputs. For text or category data, resilience is built into the system, but the approach differs from traditional software development.
“People are used to ‘we build it, it works, we forget about it,’” Rose said. “That’s just not how these systems work.”
Practical Example from the Medical Sector
Rose shared a case involving a medical client migrating to a new billing reconciliation system. Records existed in mixed formats: PDFs, images, and scanned documents. Information was often inconsistently placed, with doctor names appearing in patient fields and vice versa.
The generative AI was able to extract clean data from simple prompts, applying optical character recognition to images and text extraction to PDFs. More advanced agentic approaches then compared customer records against insurance contracts to verify billing rates.
“You start to layer different use cases on top of one another,” Rose explained. “That’s not to say that it gets everything right. You still need a human in the loop. But what you want to do is say, ‘we started at 20% automated, and then 40%, and then 60, 80%,’ and kind of grow that over time.”
Focus Shifts to Cost and Portability
Looking ahead, Rose expects the conversation around AI models to move from raw capability toward cost sustainability and portability. He predicts a shift away from radical leaps in model performance toward making these systems viable outside large data centers.
“The last mile is how do we get these things to run on a laptop or a phone instead of having to run in a data center?” Rose said. He noted that models have already been trained on vast datasets, including much of the internet. “It’s not like there’s a tonne more data that hasn’t already been put into them that’s going to lead to some type of breakthrough.”
This focus on reducing infrastructure demands aligns with broader industry efforts to make AI more accessible and less energy intensive.
Questioning SaaS Vendor Dependency
Rose also offered a controversial opinion: organizations should consider building AI systems internally rather than buying from software-as-a-service vendors. He argued that most companies already have a cloud presence and can begin implementing agentic workloads without new software licenses or training.
“Almost everybody’s got some kind of cloud presence, and that’s where I would start,” Rose said. “Because the cloud tooling, especially for the big three, has everything you need to start implementing agentic workloads tomorrow.”
He acknowledged that the process is not overly difficult, though organizations may need guidance for subsequent steps. The approach, he suggested, can reduce long-term costs and increase control.
As these discussions continue at industry events such as the AI & Big Data Expo, where JBS Dev is participating, the focus is likely to remain on practical implementation and sustainable scaling. Rose expects ongoing debate over how to balance model capability against operational costs, with portability emerging as a key priority for the next phase of AI adoption.
