In Stay Sharp Episode 131, The AI Reality Gap – What You Need to Know Before Building at Scale, co- hosts Jonathan Scott and Juliann Grant welcome returning guest Diego Tamburini, PhD, AI Practice Director and Executive Consultant at CIMdata. Diego previously joined the podcast in Episode 125, Strategic Patterns for Implementing AI in Manufacturing, to describe five patterns of AI use and implementation in the manufacturing industry, and he returns to discuss how theory and shiny demos meet practice when implementing AI.
The Full Spectrum of Implementations
Diego summarizes the range of AI implementations that exist, starting with AI fully embedded in solutions you buy—or already have—which don’t require more work on your part. You just use it. The next level of complexity is low- or no-code work, where “You do something and the tool does a lot for you,” he says. These are typically the middle-ground implementations, where you’re using an AI-development tool to configure a retrieval-augmented generation (RAG) environment that will get data from multiple trusted sources of data and use it to generate responses. The last resort, and the highest level of complexity, is to develop an AI solution from scratch—but only if you have to.
Diego Tamburini, PhD
He makes the point that you shouldn’t try to reinvent the wheel—at least not at first. “Go as far as you can with the stuff you have,” he says. “And then when you hit the limit, which is more likely when you need to do stuff across silos, that’s where you have to start thinking I’ll probably need to develop my own RAG solution or my own agentic AI applications.” An AI agent is actually a software application that uses a large language model (LLM) and the tools it has available to achieve a goal. Jonathan emphasizes that point. “It’s still software. There’s still a design to it, there’s still an architecture to it. You need to think about the components, what tools can it use. But you have to think about it differently.”
The Skills Required for Implementing AI
Any way companies implement AI is likely to necessitate new or enhanced skillsets or capabilities in their workforce. Prompt engineering, meaning knowing how to work with AI, is the first skill Diego says every knowledge worker will have to know. After that, they need to generally understand how AI works—what an LLM is and how it operates. Or maybe some data and integration skills, especially if you’re developing a RAG application and need to connect PLM, ERP, and other systems.
He says, “Something really interesting is happening here that engineers with subject matter expertise and scientists will be expected to be able to develop and curate their own agents. They have the subject matter expertise, and now, using some of the low code or no code tools, they don’t have to worry about coding, deployment, or management of the agents. They can focus on capturing their expertise in an agent.” Jonathan agrees that the most straightforward path seems to be teaching the expert—an engineer or SME—to curate the AI agent, rather than asking IT to become SMEs, but that’s going to require those SMEs to become more familiar with how agentic AI operates.
Implementation Challenges
Diego mentions the challenges of navigating through the mismatch between slick demos and what gets deployed, as well as the conflation that occurs when everything is “AI,” even if it isn’t or shouldn’t be. But he also explains that companies already struggling with the “make, buy” decision really need to adding a “wait” option. Waiting is relevant to implementation, because when every provider is adding AI to their solutions, they may be adding something you need faster than you can build it. So take a moment and investigate what’s out there. But he notes, “Don’t wait to start getting acquainted with AI. That cannot wait. You have to start looking at it. You have to start seeing are there any pain points in my company that I should be addressing with AI? That kind of thinking needs to be done now.”
Learn More About AI Implementations
The full podcast contains more discussion and details on a variety of topics, including why organizations want to have a dedicated layer in their AI model for the digital thread, what Diego says is the best way to become good at working with AI, why he says engineers especially will always need to be in the loop with AI, which primary security risks to watch out for when curating your own AI agent, and more.
Be sure to check out Stay Sharp Episode 131: The AI Reality Gap – What You Need to Know Before Building at Scale and join us each week for a new podcast.
Follow the Razorleaf Podcast, Stay Sharp in Digital Engineering on:
