AI has gotten really smart. The latest models from OpenAI, Anthropic, Google, xAI, and others can reason through complex problems, analyze enormous amounts of information, write code, create documents, manipulate spreadsheets, search the web, and increasingly operate software on our behalf. And the models keep getting better. Yet ask one of those same models to perform real work inside a commercial real estate firm and you quickly run into a problem. The AI is smart. It just doesn't know commercial real estate. It doesn't know your underwriting methodology. It doesn't know how an experienced acquisitions professional approaches a T12. It doesn't know which fields matter when abstracting a lease. It doesn't necessarily know the difference between how you analyze a multifamily rent roll and an industrial NNN rent roll. And even when it understands the terminology, it usually doesn't have access to the specialized data required to actually do the work. That gap between general intelligence and domain expertise is increasingly becoming a context problem . And one emerging solution to that problem is Context as a Service , or CaaS. The Problem Isn't That AI Isn't Smart Enough For the past several years, much of the conversation around AI adoption has focused on the models themselves. Which model is smartest? Should I use Claude or ChatGPT? Which model has the largest context window? Who scored highest on the latest benchmark? Those questions matter. But increasingly, I think they're becoming less important. The frontier AI labs are engaged in a massive arms race. Models are improving quickly, and the differences between them will continue to change. The model that is "best" at a given task today may not be the best six months from now. Meanwhile, a different problem remains. A general-purpose AI model has been trained on an enormous amount of information about the world. But it hasn't spent 20 years working in acquisitions. It hasn't built hundreds of institutional real estate models. It doesn't have your firm's investment committee process memorized. And it doesn't automatically have access to the property, market, demographic, ownership, zoning, rent, or transaction data that CRE professionals use every day. In other words, intelligence isn't the same thing as expertise . This is becoming an important distinction across enterprise AI. Anthropic describes "context engineering" as the natural progression beyond prompt engineering. Rather than focusing simply on how a prompt is worded, context engineering asks a broader question: what information, instructions, tools, data, and history should the model have available when completing a task? That is a much more useful way to think about AI in commercial real estate. What Is Context in AI? When we talk about context, we're talking about the information an AI has available when it reasons through a problem. That context can include things such as: Instructions describing how a task should be performed Methodologies developed by experienced practitioners Reference documents Company-specific standards and preferences Market and property data Previous conversations or decisions Software tools the AI can use Examples of what a good finished deliverable looks like Think about how we train a new analyst. We don't just hire a smart person, hand them Excel, and say, "Good luck." We teach them how we underwrite. We give them templates. We explain what goes into an IC memo. We show them previous deals. We give them access to CoStar, RealPage, public records, internal databases, or whatever other information they need. We teach them which assumptions matter, which numbers need to reconcile, and where mistakes tend to hide. Over time, that analyst combines their underlying intelligence with institutional knowledge, data, processes, and experience. AI needs the digital equivalent. From Prompt Engineering to Context Engineering The first generation of generative AI adoption placed a lot of responsibility on the user. If you wanted a strong result, you had to become good at prompt engineering. Instead of: "Underwrite this deal." You learned to write something more like: "You are a senior acquisitions professional with 15 years of experience underwriting institutional multifamily..." Then you'd explain the methodology, specify the inputs, define the output format, attach supporting documents, tell the model what assumptions to use, and perhaps provide examples. With enough effort, you could get a pretty impressive result. But there's an obvious problem with that approach. Most commercial real estate professionals don't want to become AI engineers. And they shouldn't have to. The acquisitions professional should be focused on finding and evaluating investments. The broker should be focused on winning assignments and closing deals. The asset manager should be focused on maximizing asset value. Those professionals possess incredibly valuable domain expertise. Asking each of them to in