An interesting theme has been developing for much of this year about how AI is changing how leaders think about the technology they are using.
For a number of years, AI has promised to sweep everything before it. In tech terms, it probably will, but we are clearly in the early innings of that transformation in multifamily. Expectations remain relatively immature, and while AI is already ubiquitous in multifamily technology stacks, it remains far from clear what AI is actually supposed to do for operators.
It is a thread that I plan to pull on in multiple forums over the coming months, as I explain below.
The LLM Boundary
For the last few years, the 20for20 Annual Survey has charted AI adoption. This year’s edition noted the explosion in functional scope as AI is being deployed across an ever-growing set of prospect and resident interactions. At the same time, it is clear that adoption, and indeed the framing that operators apply to AI, are highly vendor-led. AI, like most technology, is considered in the context of specific solutions. But the scope of AI keeps changing, so the bigger question of “how should AI work for us?” matters.
An increasingly popular opinion about technology is "I can just get Claude to do this." There is no doubt that the ability to vibe-code solutions rewrites many of the rules of software. But the opinion also suggests that we may not yet understand the boundary between immensely powerful large language models (LLMs) and the multifamily software stack.
I was reminded of this at NAA Apartmentalize, where a session presented invoice processing as something Claude could perform more efficiently than people. That is probably true: Claude can almost certainly review invoices more efficiently than people can, but that doesn’t necessarily make it the right tool for this job.
Invoice review is a highly formulaic, predictable process, but Claude is a non-deterministic tool, which means the same prompt can produce different answers from the same data. Further, while it can save a great deal of time, it may not save money. If every invoice requires a separate API call to a large language model, the cost of the process may prove to be much higher than it first appears.
The Claude take above suggests that anything a software provider can do can just as easily be vibe-coded in Claude. But that misses an important point. The most capable software companies are already using Claude to develop their proprietary software. There is a place for LLMs and a place for proprietary software. It’s worth putting more definition on where the boundary lies, even as things continue to change.
Competing With AI-First Technology
Much of the discussion at NAA recently centered on the forthcoming disruption to the property management system (PMS) market. As AI takes on a larger role in operations, it is changing what operators need from software and redefining competition among established vendors.
Another important finding from this year's 20for20 Annual Survey was that some of the industry's more sophisticated technology leaders are beginning to evaluate software based on whether AI is sufficiently central to its architecture for the product to remain viable over the long term.
This points to the concept of AI-first or AI-native architecture, where AI is central to the architecture of the software. Either it was built that way from the outset, or it has been re-architected so extensively that it is equipped to compete over the long term. That is quite different from “AI-enabled” software, i.e., where AI features have been added to non-AI-native software.
Where to Learn MORE!
Understanding that distinction is essential, and it is a topic I will explore in detail through this fall. First, it will be the subject of a brand new series of 20for20 podcast conversations, which is currently in production.
The pod series will dive deep into the concept of AI-first and what it means to technology leaders and the founders of the vanguard of leading AI-first technologies. The use cases will run the gamut from PMS technologies to category-leading best-of-breed AI.
And to ensure that we are getting deep enough into the critical question of “Could Claude do this for us?” I will be joined for each interview by Jonathan Buckelew, founder of the CRE AI Studio, who is one of the foremost experts on applying LLMs to commercial real estate.To subscribe to the season, search for Tech Talk with 20for20 wherever you get your podcasts, or subscribe to our new YouTube channel.
And at the same time, I will be covering this topic in person at Blueprint this September. My session, "What Does AI-First Have to Do With NOI?", will examine three case studies covering development, asset management, and operations. Each will explore what AI-first technology makes possible that conventional software cannot.
I hope as many of you as possible will join me for both sets of discussions - you can register for Blueprint and access a $300 registration discount at http://blueprintvegas.com/20for20. I will publish more details on that session on this blog in the coming weeks.
