AI Output Is Becoming Cheap. Operational Value Isn’t.

[Georgii Grigoriants]

Over the last few weeks, I’ve noticed more and more investors talking about AI agents, automated underwriting tools, AI workflows, and “fully automated” deal analysis.

Some of it is genuinely useful.

But real estate may be entering the same phase the software world is starting to encounter now:

AI systems that generate enormous amounts of activity without creating proportional operational value.

An AI tool that reduces underwriting time from 8 hours to 2 hours is valuable.

An AI tool that generates beautiful reports, endless summaries, and impressive-looking analysis — while the investor still manually verifies everything afterward — may simply become expensive operational noise.

The real question is no longer:
“Can AI generate output?”

The real question is:
“What measurable outcome actually improved?”

Did the system:

• reduce underwriting time?
• identify risks earlier?
• improve screening quality?
• accelerate decision-making?
• reduce expensive mistakes?

Or did it simply produce more analysis?

As AI becomes increasingly embedded into real estate workflows, the most valuable systems probably won’t be the ones generating the most output.

They’ll be the ones most tightly connected to real operational outcomes.

[Chris Seveney]

While output may be easier, the important question is: what is the quality of that output. Just like any system technology model evaluation, whatever it is you're doing, the same philosophy applies, which is "garbage in, garbage out." What appears to be happening today is that many people are using ChatGPT or other types of language models that just have data from everywhere on the Internet, which is good and bad. The issue is that there's so much garbage mixed in. The output is not truly defined to the individual who is using it, and what you end up with is something no different than, for example, people buying leads or skip tracing. It's a software and system that people go and use and input the same information and get the same output. That does nothing to differentiate you from everybody else. The question should be: if you're using this type of technology, how are we using it to differentiate yourself?