
Since bursting on the scene in late 2022, few topics continue to capture more attention than Artificial Intelligence. I first wrote about AI in this very column in the March 2023 edition, with a follow up in November 2024. Given the extraordinary pace of evolution, it’s about time for another update. Over the past 18 months, several trends have emerged: more capable reasoning models, autonomous AI agents, multimodal systems, specialized industry models, and growing adoption of private large language models (LLMs). Together, these developments are reshaping not just the commercial real estate industry but also how general businesses operate and how knowledge work is performed.
Give me another reason. One of the most significant advances in AI has been the shift from simple, prediction-based language models to systems capable of deeper reasoning. Leading AI providers have focused on improving models’ ability to solve complex problems, write software, analyze data, and execute multi-step tasks with greater reliability. OpenAI, who started the AI party with GPT-3+, released GPT-5 in August 2025. This update introduced a unified architecture that can determine when a task requires deeper reasoning with slower response versus a simpler but faster response. It also includes major improvements in coding, writing, scientific analysis, and instruction-following, while also reducing hallucinations and factual errors. At the same time, the broader AI industry has increasingly emphasized “reasoning-first” training techniques, including reinforcement learning approaches that reward verifiable outcomes rather than merely predicting likely text. Researchers and industry analysts view these methods as essential for improving performance in high-stakes domains such as law, finance, healthcare, and engineering. Claude ai has become a very popular product in these sectors specifically because of its advanced reasoning models.
Not so secret agents. Another major development is the rise of AI agents. Unlike traditional chatbots that simply answer questions, agents can perform tasks, use tools, access data sources, and execute multi-step workflows autonomously, giving rise to a new term – agentic AI. Enterprise software vendors are rapidly integrating agents into business operations. Microsoft, for example, has expanded its Copilot product with specialized agents such as Researcher and Analyst, while introducing multi-agent coordination that allows multiple AI systems to work together on complex processes under human supervision. AI agents have become built into specific layers to help weave ai automation within the human structure of the organization. For example, an AI agent may help organize and summarize the results market research study, which is then handed off to a financial analysis model to help form and validate input assumptions, which then loops in a construction management study to help determine the project’s viability. Rather than replacing employees, these agents are increasingly being positioned as digital coworkers that augment human productivity across a range of tasks.
A picture is worth 1,000 words. While early AI systems were limited to only recognizing text, modern AI systems have evolved to be able to process and generate information across multiple formats, including images, photographs, audio, documents, and video. This multimodal capability allows users to upload presentations, analyze diagrams, interpret spreadsheets, generate visual content, and interact naturally through voice. Multimodality is considered one of the defining characteristics of next-generation AI systems because it more closely mirrors how humans gather and process information. For businesses, multimodal AI dramatically expands practical applications. Employees can ask questions about contracts, analyze photographs from inspections, review technical drawings, summarize meeting recordings, and generate polished reports from diverse information sources with a single system.
Bigger is not necessarily better. While large, general-purpose models remain important, organizations are increasingly looking to specialized and vertical AI models optimized for specific industries and use cases. Rather than relying entirely on one universal model, many enterprises are deploying domain-specific LLMs trained on proprietary datasets. Specialized models can offer higher accuracy, lower operating costs, improved compliance, and better performance within narrow domains such as such as a specific type of healthcare, such as cardiovascular or a targeted legal service, such as litigation. Small Language Models (SLMs) are also gaining popularity. These models require fewer computing resources while delivering strong performance again by focusing on targeted applications. Their lower infrastructure costs make them attractive for organizations seeking scalability without the expense of running frontier-scale models.
Going private. Perhaps the most important trend for enterprises is the growing adoption of private LLMs. A private LLM is a language model used exclusively within an organization’s secure environment rather than accessed solely through a public cloud service. These systems may run onsite, within a private cloud, or inside dedicated enterprise AI platforms, such as Google Notebook LM. Private models allow businesses to retain control over their data, intellectual property, security policies, and compliance requirements. Several factors are driving this trend. First, organizations increasingly want assurances that proprietary information will not be used to train public models. Industries such as healthcare, banking, government, defense, and legal services often face strict regulatory requirements around data privacy and sovereignty so a private LLM allows them to harness the power of AI without introducing widespread security concerns. Second, enterprises are looking to customize AI systems using internal knowledge and business processes. Microsoft’s Copilot Tuning initiative, for example, enables organizations to tailor AI models with company-specific workflows and information while keeping operations within Microsoft’s protected and highly secured enterprise environment. Microsoft also emphasizes that customer data is not used to train foundation models. Third, advances in model efficiency have made private deployment more practical. Open-weight and smaller models increasingly deliver capabilities that were previously available only through massive cloud-hosted systems. This allows organizations to balance performance, cost, and control more effectively.
The future of artificial intelligence is likely to be defined less by large-scale models and more by smarter, more targeted applications. Organizations are increasingly combining reasoning-capable AI, autonomous agents, multimodal interfaces, and private LLMs into their specific business systems. The next phase of AI adoption will focus on trust, governance, security, and measurable business outcomes. Private LLMs, in particular, are emerging as a critical component of enterprise AI strategies because they allow companies to harness the power of generative AI while maintaining control over sensitive information. It all circles back to a phrase that I used in my original article a few years ago – AI is not going take your job, but the person/company that knows how to efficiently use it just might!
What IC @ PVC – Testing the market. Last month, a pair of familiar Rockside Road office buildings were put up for sale. Rockside Square One and Two, encompassing a total of 156,000 square feet, has an asking price of $18.35 million or $118 per square foot. While the complex features a strong location, solid occupancy and is in good condition, the market will be closely eyeing how this sale effort unfolds.
Alec Pacella, CCIM for August 2026 Properties Magazine




















































