|
This year is supposedly the “year of the AI Agent”. You will be hearing a lot about agents this year. Basically, they attempt to replicate human actions to go do stuff, instead of just having a conversation with you. Someone once told me “tech is about maximizing laziness.” Sending an agent off to do stuff for me while I doom scroll on Eagles chatter work productively is definitely maximizing laziness…perfect! ChatGPT released its agent, Operator, Thursday night. I messed around with it on Friday and was pretty impressed. I started by having it look for Eagles tix for Sunday's NFC Championship game - go Birds! (Sidebar: I started off as a journalism major and hate how my kids text without capitalizing words…and yet I find myself doing the same now…”so cringe” they would say) Me: find me some good tickets for the eagles nfc championship game on Sunday Operator: I found tickets for the Eagles NFC Championship game on SeatGeek, but it asks for the number of tickets. How many tickets would you like? Me: 4 Operator: There are 1,898 ticket listings for the Eagles NFC Championship game, with prices ranging from $657 to $3,871 each. Do you have a specific budget or seating preference in mind? Me: start with the best seats possible. I like being as close to the 50 yard line as possible, lower bowl but not in the first 15 rows of seats because it's too close. Operator: I found tickets in Section 211, Row 15, for $657 each. These seats are in the lower bowl and offer a good view of the field. You can view these tickets here. Would you like to proceed with these? Me: show me the view from those seats first Operator: Here's the view from Section 211, Row 17 at Lincoln Financial Field. It provides a clear perspective of the field. Let me know if you'd like to proceed with these seats or explore other options. |
|
Me: i said I want to be close to the 50 yard line, not the endzones Operator: I found tickets near the 50-yard line in Section 101, Row 18, which is in the lower bowl but not in the first 15 rows. The view from these seats offers a great perspective of the field. Would you like me to proceed with these tickets? I stopped there, but had I continued it would have bought the tickets for me. It also showed me searches across different ticket vendors. Pretty cool, but not exactly revolutionary. Also, it was slow. I tolerate that because I can send it off to work while I do something else, but I wouldn’t do this if I was in a hurry. Let me try something else… What if prospective tenants start looking for apartments this way? Here’s a screen capture video of what my first pass looks like. This is just meant to give you a feel for what the experience is like, not to suggest I know how to do this or that this is better than doing it myself. Warning: this is 3 minutes long, but I wanted you to see it in action. |
|
One of the first things that stands out to me is that it might be better suited for a very broad search rather than something with specific filters you already have in mind. This is particularly true if the website allows for saveable filters. It’s not that hard to go to apartments.com and save your filters. Another thing that jumps out at me is that websites might need to be optimized for agent searches. This was painfully obvious using our own website. I asked it to get me a defeasance quote from Pensford and it struggled with calendar drop downs. In this video, you can see that it gives me the option of taking control, choosing the calendar date, and then giving control back. That’s great, but why not just do it myself if that’s the case? Warning: this one is about 90 seconds. |
|
When I asked it to get a cap quote from Pensford’s website, it said the Cap Pricer Excel download was blocked. Does that mean we need to change the Pensford website to allow for the ability to type in cap terms right on the website instead of a download? I don’t know, but I do wonder how much energy will be spent optimizing websites this year. There was one task that was hugely impactful for me - abstracting property addresses from borrower websites. We run addresses through our proprietary algorithm and match properties with Freddie/Fannie/CLO/CMBS publicly available loan information. This gives us real time insight into borrower’s portfolios, even if we aren’t working with them. And for our broker clients, it helps them keep tabs on their clients’ deals with competitors (“Hey, that deal you did with XYZ has IO ending in 6 months, should we be talking about a refi?”). It’s actually the coolest thing we’ve ever built, but it requires a lot of manual effort to keep up with sponsor addresses. It took over an hour to abstract this borrower’s property addresses, so I will spare you the video. But it did 93 properties on its own. More impressively, the addresses weren’t listed on the borrower’s main website. It had to navigate to “Portfolio”, then click on each property, get the address, backpage, then click on the next property. Then it spit out the results so we can ingest into LoanBoss and match Agency loans to this borrower. |
|
We already do this manually for over 1,000 sponsors every quarter. This feels like it could help free up a lot of my team’s time immediately. One last thing I did was give it my login information to some aviation software I use that doesn’t contain any confidential information. I asked it to go get me the upcoming maintenance I need to have done…and it worked. Very smoothly. Is it a stretch to envision giving my login information and credit card information so it becomes my hub for most activities? Or sending it off to get a report out of Yardi? Or run projected DSCR in LoanBoss with a 50bps interest rate shock? Lots of potential here depending on how the Agent uses my information. Takeaways
That being said, I see the potential. I was impressed with how effective it was right out of the gate. It’s not better than doing it myself…yet…but I see the vision. Like I wrote last month, "This is the Worst it Will Ever Be."
|
|
|
|
|