Newsletter · · Ashutosh Agarwal
AI Starts Doing the Front Line Work of Proptech and Construction - Vertical Spotlight - Week of August 25, 2026
Startups and venture newsletter for the week of August 25, 2026, spotlighting proptech and construction, the week the software stopped assisting and started doing the job: leasing calls, appraisals, inspections, and underwriting.
Vertical Spotlight: Proptech & Construction
Week of August 25, 2026: AI Starts Doing the Front Line Work of Proptech and Construction
The week the software stopped assisting and started doing the job: leasing calls, appraisals, inspections, and underwriting. Covering proptech and construction podcasts from Aug 18–25, 2026.
The Landscape
For years, "AI in real estate" mostly meant a chatbot bolted onto a website and a lot of slides about the future. This week the podcasts described something much more concrete: AI that is actually doing the front-line work of the property business, answering the leasing phone, filling out the appraisal, walking a tenant through an inspection, and reading a deal's risks, and doing it with numbers attached.
Three threads ran through nearly every episode:
- AI is being pointed straight at the industry's labor shortage. Real estate and construction are both short of people: leasing agents to answer calls, appraisers (many near retirement), and skilled trades to actually build things. Almost every builder this week was using AI not to add a feature, but to cover work there simply aren't enough humans to do. One co-living operator said a single salesperson went from handling about 7 leads a day to 40–50 once AI took over the back-and-forth, an 8x jump in how much one person can process.
- The metrics finally showed up. This wasn't hand-waving. An appraisal startup is turning a 3–4 hour report into a 30-minute one. An apartment operator's homemade leasing bot handled 729 conversations and captured 446 leads in its early days. A co-living roll-up says it lifts a newly bought company's profits 2–4x by swapping in its own AI software. When the claims come with numbers, you can start to judge them.
- A surprising number of founders think this is the end of traditional software. More than one guest argued that the old model (buy a subscription to a tool, then have humans operate it) is dying. As the co-living operator put it bluntly: "I don't believe that there will be any software as a service in the next three to five years." The bet is that AI doesn't just help you do the task; it does the task, and you pay for the outcome.
The debate we close on is the human one, because it hung over the whole week: if the software can now make the calls, run the inspection, and write the report, what exactly are the people for? The podcasts had confident answers about the technology. They were genuinely split on the people.
Here are the companies people were actually naming.
Companies to Know
ValuMate: the 30-minute home appraisal
The clearest "before and after" of the week came from the appraisal world, which is staring down a November 2 deadline to switch to a new government report format (called UAD 3.6) that barely resembles the old one, "only 3% of the current form statically maps to the new form," as CEO Anir Pratavati put it. Most existing software makes that painful: appraisers told ValuMate the current tools take "two and a half, three and a half, four and a half hours" to produce a report. ValuMate, founded in 2025 by three Carnegie Mellon AI researchers and part of Y Combinator's spring batch last year, says the inspection and report together "shouldn't take more than about 30 minutes", roughly a 15-minute inspection and 15 minutes for the appraiser to review. Under the hood, when you upload your files it deploys 31 different AI models, or agents, in parallel, each "trained to think like an appraiser," filling in the report while leaving the true judgment calls to the human. Latest runs take about seven minutes of processing; the inspection app uses an iPhone's LiDAR scanner and lets the appraiser just talk to it (it's smart enough not to dump your chit-chat with the homeowner into the report). Both Fannie Mae and Freddie Mac have verified ValuMate for the new format, and the founders proved out the product by running their own high-volume appraisal firm, VM Appraisals, on it. (Appraisal Buzzcast, "AI, UAD 3.6, and the 30-Minute Appraisal," Aug 19)
Gray Residential's "Remy": the leasing agent an operator built rather than bought
On a Gray Report deep-dive, the operator walked through the results of Remy, an AI leasing agent they built in-house to answer prospect and resident calls. The early data: 729 conversations, 446 leads captured, 32 of them after hours (13% of calls came in after hours), and, by their own extrapolation, "over nine hours of staff work" saved so far. The pitch is simple: "there's no more missing a call." The reason they built it themselves is a number worth remembering when you hear about AI leasing tools: off-the-shelf products, they said, run roughly $10,000 to $120,000 a year per portfolio. And crucially, they were adamant this isn't about firing the leasing team: "we're not removing our site teams... if someone's available" a human still picks up; Remy is the safety net that catches the calls people would otherwise miss. (The Gray Report Podcast, "Inside a 100-Year Flood," Aug 21)
Nebo: the co-living roll-up that buys companies and rebuilds them with AI
The most striking operational numbers came from Sergey, who runs a co-living company with 4,000 rooms and built his own software platform called Nebo. Before properly implementing AI, he said, one salesperson could handle about 7 leads a day, "their sweet spot maximum." After: 40 to 50 leads a day, which he called "close to an eight-time improvement in terms of volume." His rough split of where AI now carries the load: about 80% of the leasing and marketing workload, 60% of customer service, 60% of financial operations, and only about 10% of field operations so far. The business model is the interesting part: his firm buys traditional property-management companies and uses the software to lift their profits (EBITDA) by 2–4x, that's the first thing they check before buying. He plans to open Nebo up to smaller co-living companies near the end of 2026, and his reasoning was the week's boldest line about the industry: "I don't believe that there will be any software as a service in the next three to five years." (The Co-Living Show, "EP 28, Inside the Country's Largest Co-Living Operation," Aug 20)
Residia AI: turning the dreaded property inspection into a revenue line
Founder Peter Fife built Residia out of his own frustration running rental properties, where "inspections and maintenance have always just been a pain." His insight: property managers are supposed to inspect units two to four times a year, but the job is so awkward (nobody enjoys walking into a trashed unit and telling the tenant) that many managers "just don't do them at all," and the hidden damage surfaces at move-out. Residia flips it: the tenant does a guided self-inspection on their phone, with the app telling them to "back up or take a picture of the oven or the filter under the sink," then it generates two reports: one on the unit's condition, and one that flags maintenance work as "revenue-generating opportunities" (many managers add a 10% charge or a dispatch fee on maintenance, which is also how Residia makes money, a cut of completed maintenance invoices). It launched to property managers in August after a beta with DIY landlords, and is hitting the industry conference circuit (the property managers' association, NARPM) to scale. (The Midnight Founders Podcast, "Peter Fife, Residia AI," Aug 20)
Polly: automating the unglamorous machinery of a mortgage
On Chrisman Commentary, Polly's Brandon Story, who was previously chief investment officer at mortgage giant Mr. Cooper, laid out where AI is cutting cost inside home lending. The mortgage industry, he said, is one of the "slower adoption industries," still "mired in Excel spreadsheets" for basic pricing. Polly automates the capital-markets plumbing: the pricing engine and rate sheets, the "lock desk" (historically a person reading an email and manually locking in a loan's rate), and, launching this fall, an AI-driven hedging platform aimed squarely at two aging incumbents, QRM and Compass Analytics, that he called "really outdated." His framing of why this matters is a good lens on the whole vertical: on a mortgage, revenue is "pretty vanilla" (everyone prices off the same market) so "where there's differentiation is going to be how efficient is your operation." In other words, in a commoditized business, the cheapest operator wins, and AI is the cost lever. (Chrisman Commentary, "8.19.26 Loan Production Costs; Polly's Brandon Story on Capital Markets Tech; Long Bond Rising," Aug 19)
Dialed: an AI "second brain" for commercial real estate deals
CEO Steven Song described Dialed as "an AI-powered decision intelligence layer for real estate", a system meant to fix how acquisition and underwriting decisions actually get made, which today he says is "mostly through fragmented human judgment": someone pulls the numbers, someone calls a broker, someone who knows the neighborhood says "that block is different from the next one over," and it all gets "compressed into a memo or, more often, into a feeling." His sharpest point was about hidden risk: firms "start confusing individual judgment with institutional capability", they think the organization has an edge when really one person does, and that doesn't scale. The risk he says everyone underweights is "contextual drift", the neighborhood around a building changing faster than the model assumes (safety perception, tenant mix, zoning, policy). Dialed tries to pull those messy, unstructured signals into the decision continuously instead of quarterly. A useful data point he gave on AI economics: doing what Dialed does was about "120% more expensive" a year ago, the underlying AI is getting cheaper fast, even as total usage rises. (Tangent: Conversations with Real Estate & Tech Innovators, "Using AI to Make Better CRE Acquisition and Underwriting Decisions, with Diald CEO Steven Song," Aug 18)
xPL: building data centers in a factory because the jobsite can't keep up
The week's most important construction conversation came from a Turner Construction subsidiary called xPL, which manufactures big chunks of buildings offsite and ships them to be assembled, an answer to the data-center building frenzy. The problem it's built for is stark: the U.S. is heading for a 14% decline in its industrial-electrician workforce by 2030 while demand rises 25% over the same window. xPL's motto is "bring the work to the person. Don't take the person to the work", manufacture near where workers live (so they "get home every night to see their families") rather than dragging thousands of them to a remote mega-site. It already uses robotic welding and Lincoln "Python" steel-cutting machines in its factories, though the executive was honest that true "smart factory" automation is still a way off because "we're not a true manufacturer", they build ~200 modules for a project, then retool the whole line for the next one. The blunt takeaway, and a real contrarian claim: with traditional construction methods, the hundreds of billions in announced data-center investment simply "would never be built in the schedules that is demanded," and even xPL plus all its offsite competitors "won't be enough to keep up with the demand." (The Dynamo Show, "Can Construction Keep Up With the Data Center Boom? xPL's Answer," Aug 19)
Elise AI: the leasing-and-resident agent going after every apartment conversation
Jacob Kosior, who spent a decade in multifamily operations before joining Elise AI, described a full suite of AI agents for apartment operators: leasing AI (prospect conversations), delinquency AI (chasing unpaid rent), renewals AI, and maintenance AI. The line that captured the leap in scale: "your assistant manager on site... can only call one resident at a time who owes them money. AI, we can call an infinite number of residents at any time", enough to "call the entire building" about balances in the time a human makes one call. Elise builds its answers from an operator's own procedures and property-management system so the bot stays on-script. Kosior's advice for smaller operators was refreshingly un-salesy: start by playing with general tools like Claude or ChatGPT for tasks like market surveys, then adopt a purpose-built suite when you're ready. (Multifamily Insights, "If AI Can Handle Leasing Tasks, What Happens to Leasing Teams with Jacob Kosior, Ep. 805," Aug 19)
Nordon: a credit check for senior-housing operators
On Proptech Espresso, Jerry Vinci described Nordon, launched a couple of months ago, aimed at investors buying senior-housing communities. His observation: investors do thorough financial diligence but "completely miss the one variable that decides the outcome", whether the operator can actually sustain or grow occupancy in that specific local market. Financials are "past numbers," he noted; they don't tell you the future. Nordon runs an independent diagnostic from outside-observable signals (how visible the community is on Google and in AI search results, its reviews, its referral relationships, its pipeline for the next 6–12 months) to catch problems "that start to go wrong before the financials" do, without relying on data the seller hands you. He's also bullish on virtual staging: letting families tour any unit digitally means operators can stop keeping empty "model" apartments and "recover revenue from rooms that previously were just pure overhead." (Proptech Espresso, "Jerry Vinci, Improving Senior Living Facilities For Families & Operators," Aug 20)
Bluebeam "Max": AI reaches the small general contractor
A grounded counterweight to all the startup hype came from Melissa Drew, who runs a small construction firm and is decidedly "not on the leading edge of AI and robotics." Her team lives in Bluebeam, the drawing-markup tool, whose new AI feature "Max" she called a "total game changer." The specific win was mundane and real: Max's "stitching" feature automatically joins drawing sheets together across a set: on a marina project "spread out over five different pages," it let them see "the whole picture" at once instead of manually snipping and resizing PDFs. Combined with layers they could toggle on and off, it let them run bid sessions where subcontractors could suddenly grasp the whole project, spotting where "the mechanical guy was going to cross with sewer pipe." Her broader use of AI is equally practical: feeding "thousand-page spec books and thick contracts" into tools like Claude and ChatGPT to speed up risk assessment on which jobs to bid. (The ConTechCrew, "The Construction Project Mistakes... Builders Destroy Profit by Rushing Work with Melissa Drew," Aug 18)
One Debate: If the AI Makes the Calls, What Are the People For?
The question was literally in an episode title this week, "If AI Can Handle Leasing Tasks, What Happens to Leasing Teams?", and it echoed across the whole vertical, because this is the week AI stopped being a helper and started doing the actual jobs people are paid to do.
There are three distinct answers floating around the podcasts, and they don't agree.
Answer one: nobody gets fired; the humans just get freed up. This was the comforting, most-repeated version. The Gray Residential team was emphatic that Remy is a backstop, not a replacement: "we're not removing our site teams"; a human still takes the call when one is available. The co-living operator, Sergey, went further and reframed it as a lifestyle upgrade for his commission-based salespeople: with AI doing the grunt work, a person can "work two hours instead of eight and earn whatever you earn now, or work six, eight hours and earn four times more." In his telling, "AI will not take over many jobs, but instead it will allow people to choose what they want to do."
Answer two: the human's job changes into something harder, asking the right questions. Dialed's Steven Song argued the valuable human skill in this new world is judgment and framing, not doing the legwork. His system deliberately won't answer off-topic prompts (they tested asking it "what should I have for lunch" and it stayed on task), because the discipline is in "asking the right questions so you don't burn tokens asking the wrong question." The person who thrives isn't the one who can pull the data (the AI does that), it's the one who knows which question actually matters.
Answer three, unspoken but hanging in the numbers: maybe you just need far fewer people. Nobody said this out loud, but the math is hard to ignore. If one salesperson now does the work of eight (co-living), if 60% of customer service and 60% of finance work is automated, if an appraiser's 4-hour report becomes 30 minutes, and if a leasing bot can "call the entire building" at once, then even in a world with a genuine labor shortage, the number of humans each operator needs is falling fast. The optimists are betting the shortage is so severe that AI just fills the gap and nobody loses a seat. The quieter possibility is that the shortage is exactly what makes it socially acceptable to automate these roles now, before anyone has to decide whether they'd have hired those people anyway.
Neither the industry nor this week's guests resolved it. What's striking is that the people building these tools mostly land on answer one, and keep publishing metrics that make answer three look inevitable. That tension is the thing to watch in proptech and construction: not whether AI can do the work (this week made clear it can), but who's still standing on the other side of the productivity curve.