Samuel Holcman: Thank you for tuning in to Real Talk. Be sure to join your host Sam Holcman again for another edition of our program. We'll have more Real Topics of discussion then. How do you sort out the so-called jargon from Enterprise Architecture and Business Architecture practices that work? Do the members of your organization find some business or technology advice utterly confusing? Welcome to Real with Sam Holcman. ⁓ In this program, we the record straight ⁓ and in terms that business people and technology people can understand. Now, here is host, Sam Holcman. Rethinking Enterprise AI. Why small models fit big organizations. This is Real Talk with Sam Holcman. I'm Sam Holcman. Over the past few weeks, I've seen more mainstream coverage arguing that the future of Enterprise AI is not gigantic or multiple large language models, but a coordinated ecosystem of smaller, task-specific, knowledge-specific models. One recent InfoWorld piece, titled Small Language Models, Rethinking Enterprise AI Architecture, makes this case clearly. As large language models hit its limits of scale and cost, as the industry is now calling it, small language models are merging as the faster, cheaper, more private workhorses for the enterprise. The article highlights three big advantages of small language models. A better division of labor between small and large language models, dramatic cost savings for high volume workloads, and most importantly, stronger privacy when models can run on premises or at the edge. It also points out that real differentiators is not the model itself, but the architecture around it. Data preparation, data governance, and the way multiple models are orchestrated across workflows. Now, why am I smiling? That framing will sound very familiar to anyone who has worked with the enterprise architecture center of excellence, EACoE. or the Business Architecture Center of Excellence, BACoE. Over the past, yes, decades, for more than 10 years, especially recently, we have been building and refining what we call EAI. Those initials have three different versions that our clients have used. Enterprise augmented information, enterprise artificial intelligence, and enterprise amalgamated information. The core idea has never been one model to rule all of them, but an enterprise architecture in which AI is composed of many focused capabilities, each grounded in well-defined information decisions and outcomes that are your best practices that are never ever in a large language model anyways. In practice, this means start with decisions that matter most, design the information, semantics, and governance needed to support them, then bring the smallest, most appropriate AI capabilities, multiple smaller domain-specific models, rather than a single generalist. A recent EACoE-BACoE engagement with a global service organization is a good example of this. Their help desk was struggling with long response times and inconsistent routing across numerous ticket categories. Exactly the kind of high volume, predictive work that a small language model, EAI, is well suited for. Instead of deploying single generalized chat bot, we used the EAI principles to model the service domains to find the information needed for Criage and then to introduce a small domain specific model that are our, their best practices. the way their clients really see competitive advantage to classify and route tickets. The result, better, faster, more consistent routing, lower cloud costs, and better use of human expertise when required while keeping sensitive operational data, client data, and internal best practices within the organizational control. The InfoWorld article also describes this pattern more broadly. Small language models that excel at narrow, repetitive, latency-sensitive work like customer service triage, ticket classification, and clause identification. Cause identification. All these things often outperform generalist large language models in these zones and reduce hallucinations and what we like to call fabrications. and the risk associated with those by staying closer to curated enterprise data. This is almost a textbook description of how we position EAI and its components inside of enterprise architecture and business architecture. Tightly scoped, context aware, and aligned to specific decisions and processes. Perhaps the most important point in the article is that small language models are not going to replace large language models. Actually, they're quite different. The future is composite AI, orchestrating multiple models of different sizes and deployment patterns aligned to business workflows and risk profiles. Gartner predicts that by 2027, not too long from now, organizations will use small task-specific models three times more than general purpose large language models that just have General information. No information at all about best practices. No information that is competitive from a standpoint of competition that you're facing. And of course, large language models are so suspect to all sorts of manipulation. And we know this. That's exactly where enterprise architecture and business architecture becomes indispensable. EAI gives you the structure to decide where small models make sense and where large ones still may add value in your organization. It ensures that data is prepared, governed, and visioned so that the models remain trustworthy over time. It integrates AI capabilities into real processes, not just pilots and proof of concepts. Now, for organizations that have been with the EACoE and BACoE, And on that journey, the current small language model conversation is less of a revelation and more of, and we're very happy about this, validation. You've already seen how an architecture-first approach to AI, augmented information, artificial intelligence, amalgamated information, creates doable compounding advantages. If your enterprise is now asking, how do we blend large language models, small language models, and everything in between into something coherent, that is exactly the question that EAI was built to answer from its inception. So please, please reach out to us at www.eacoe.org and www.bacoe.org to gain the competitive advantage your organization deserves. It will be our pleasure to speak with you. We look forward to hearing from you.