Key takeaways
- Most firms deploy agents as point solutions whose output ends in a chat window or a spreadsheet. Treating agents as a new class of user, with a system of record, shared definitions, permissions and orchestration, is the approach that scales.
- Adding AI agents is itself a form of scaling. A small team plus its agents needs the shared definitions, or semantic layer, of a much larger organization.
- Adoption depends on trust, and trust depends on outputs that are consistent, traceable to a source and aligned across teams. Structured data governance is what makes that possible.
- Building software has never been easier, but focus still has a cost. Buy end-to-end solutions from expert vendors where they exist, treat internal builds as temporary, and invest in data infrastructure with experienced partners.
Earlier this week, our CEO Raj Singh joined L.D. Salmanson of Cherre (now part of RealPage) for a Commercial Observer webinar, hosted by Edward Cohen, titled “AI & Real Estate: You’ve Hired an Alien Superintelligence. Now What?” The conversation centered on a question we hear from nearly every large firm we work with: once you’ve decided to put AI agents to work, how do you do it in a way that produces results people will actually use and trust?
Two ways to put agents to work
Most firms are deploying agents as a steady stream of point solutions: a skill that pulls data from an OM, another that drafts an IC memo, another that analyzes a rent roll. These are quick to build and generally do one job well enough, but their outputs tend to end up in a chat window or a spreadsheet, and rarely go further.
A less common approach treats agents as a new class of user and gives them what any user needs before she can do real work: a system of record to read from and write back to, a set of shared definitions that keep tasks and teams consistent, permissions that define what she may see, and an orchestration layer that lets one agent’s output become the next agent’s input. We believe the second approach is the one that scales, and much of the webinar explored why.
From doing the work to directing it
Raj opened with a simple observation. Real estate is a services industry, and most of the value firms add comes from people at computers processing information and making decisions. AI goes directly to the center of that. Intelligence and mental bandwidth have always been a constraint, and they can now be added quickly, cheaply and almost without limit. As that constraint loosens, the professional’s job shifts from doing the work to directing it, or in Raj’s words, from playing an instrument to conducting an orchestra.
An orchestra, of course, depends on everyone reading from the same score.
Shared definitions are where trust is won or lost
Much of the discussion focused on what Raj and L.D. described as the semantic layer (also known as an ontology), which in practical terms means writing down what your terms mean. Raj pointed out that a team of three can maintain a shared understanding informally, but a team of fifteen or twenty cannot. He also made the case that adding AI agents to a workflow is itself a form of scaling. A small group that now consists of a few people and all of their agents needs to start thinking of itself as a much larger group, with all the complexity that implies.
L.D. offered an example that will be familiar to anyone who has worked across offices. A firm might calculate net effective square footage one way in Chicago and another way in New York, often for good reasons. If two teams use the same term to mean different things, their calculations will be skewed even though the underlying data hasn’t changed. The same applies to the large language models reading that data, which have no way of knowing which definition is meant unless someone has told them.
This is the point we most want large firms to take from the conversation. An analyst who receives two different answers to what looks like the same question from two different agents will (reasonably) stop relying on either. Adoption depends on trust, and trust depends on outputs that are consistent, traceable to a source, and consistent across teams. Structured data governance is what makes that possible, and without it even a very capable agent will struggle to earn a place in the workflow.
Build, buy, and the cost of focus
The discussion also turned to a question every firm is working through: what to build and what to buy. Raj noted that nearly everyone he talks to is rethinking their tech stack, and that because it has become so much easier to build things, staying focused takes more discipline, not less. He described the “dopamine effect” of building with AI, citing a CEO who spent a few weekends vibe coding and realized he had been neglecting his day job. Something being cheaper and easier to build does not mean the cost is zero, and prioritization still matters.
L.D. made a related point: the bar to build software has never been lower, but the bar to build software that is differentiated and worthwhile has never been higher. His advice was to map the tasks and information flows across the organization and then look for an expert vendor that can address them end to end. If a problem has no solution in the market, build one, but treat it as temporary and retire it when a better option arrives, because most firms are not software companies. The one area he suggested building internally is data infrastructure, and even there he recommended working with experienced partners. Firms that believe they are saving money by going it alone, in his view, risk giving up the infrastructure they will need to compete.
Raj added that the learning curve for AI may be the steepest he has seen. Progress is moving too quickly for most teams to make mistakes they can learn from, which is a strong argument for leaving the build to those who work on it full time. The economies of scale that have always existed for software vendors being able to build more efficiently as they solve similar problems for many firms, still exists today as software vendors are also building with AI.
What this means for large firms
A single team can build a useful agent over a weekend. It is considerably harder to build and maintain the structured data layer that ensures the agent’s output agrees with every other team’s, respects who is permitted to see what, and flows into the next step of the process. In a large organization, with many teams, many agents and many legitimate variations in how terms are used, that governance is the part that determines whether AI becomes part of how the firm works or remains a collection of experiments.
That is the thinking behind how we build Altrio: a system of record for deal and portfolio data that gives people and agents a common foundation to work from. We believe the firms that get the most from AI over the next several years will be those that put this foundation in place first, with a partner that has done it before, and then give their people the tools to direct the work with confidence.
You can watch the full webinar here or read Commercial Observer’s recap here. If you’d like to talk through what this looks like for your firm, we’d welcome the conversation.