The landscape · Real estate
Real estate: what AI is really doing
Brief me in 2 minutes
AI in Real estate.
AI in property is real where it answers renters and reads paperwork: EliseAI, which says it powers one in six U.S. apartments, raised $350 million at a $4 billion valuation to run more of the back office, and PYMNTS reported in May that TD's mortgage agent had cut the wait before an underwriter could open a file from roughly 15 hours to under three minutes. The live fight is over who sets the rent: a federal judge temporarily blocked part of New York's law restricting algorithmic rent-setting, finding RealPage marginally likely to win its challenge, while a North Carolina court refused to let RealPage and its landlord customers out of the states' antitrust case. Physical work and deal decisions are still early: in August Bedrock said its driverless excavators were digging on three named projects, and in JLL's survey only 5% of occupiers said they had hit all their AI goals.
The game: we show you 6 things happening. You call each one real now, early or mostly hype. Then the receipts.
Shift 1 of 6
Leasing AI graduates from answering the phone to acting inside the property system
Real now
EliseAI says it passed $200 million in annual recurring revenue and powers one in six U.S. apartments, and investors led by Andreessen Horowitz and Bessemer paid a $4 billion price for that; Apollo itself launched in September with no measured results.
How it works. EliseAI's agents started by answering prospects and residents by text, email and phone, booking tours and logging maintenance requests. Its new agent, Apollo, sits inside the platform and inherits each user's existing permissions, so it can act on the property data directly: a leasing agent can ask in plain English for a follow-up to everyone who toured, and the messages go out.
Overstated: 'Any task on the platform' is EliseAI's own description of an agent launched in September, and the user evidence at launch was one customer calling it 'the smartest AI I know', not a measured result. It can only do what its user is already allowed to do, so it speeds up a job rather than replacing the judgment behind it.
Shift 2 of 6
Rent-setting algorithms move from a software feature to a court question
Real now
Some of the largest property managers in the country are co-defendants and the case has already produced settlements, including $7 million from LivCor; this is live legal exposure, not a forecast.
How it works. States allege that RealPage's software pools competing landlords' nonpublic rent data and recommends prices back to all of them, so rents rise together instead of competing. New York banned using such software to set rents; at the end of September a federal judge temporarily blocked the provision that bars landlords from setting rents on its recommendations, partly because it covers software built on public data as well as nonpublic data.
Overstated: Neither ruling decides the merits: the judge called it a close question and found RealPage only 'marginally' likely to succeed, and the North Carolina decision only means RealPage and the landlords have to answer the complaint.
Shift 3 of 6
Mortgage files get an AI first reader before the underwriter sees them
Early
In May one large bank had an agent live on the first step of the journey and a plan for the rest, and Copperlane, the startup doing the same, raised a seed round in June, so it works in production but is not yet how most mortgages get made.
How it works. TD's mortgage agent classifies a borrower's documents, extracts and validates the financial data, calculates income, runs policy and consent checks, flags discrepancies and delivers a summary memo before any human reviews the file; PYMNTS reported in May that at TD, pre-adjudication that had averaged roughly 15 hours took under three minutes. Copperlane's agent, Penny, does a similar first pass, scanning bank statements for large deposits that fall outside expected income.
Overstated: TD's agent routes its output to a human underwriter before any decision is made, and in May it covered only pre-adjudication: the rest of the mortgage journey was a plan TD had mapped, not a deployment. A seed-stage 'AI loan officer' is not proof that AI underwrites mortgages end to end.
Shift 4 of 6
Home search moves into chat assistants, on the portals' and the MLSs' terms
Early
These connectors are weeks old and come with press releases, not usage or conversion numbers.
How it works. Zillow launched a connector that lets Claude users search homes, see value estimates and check affordability inside the chat, a year after its ChatGPT app. The connector shows Zillow's own listing display and keeps the next step on Zillow: a tour, an agent, a loan officer.
Overstated: 'AI home search' here is the portal's own search in a chat window: Zillow says it does not send listing data to Anthropic, and every next step the connector offers, a tour, an agent, financing, runs through Zillow. There are no usage numbers yet.
Shift 5 of 6
Lease abstraction and CRE underwriting get read by AI first, humans check the flags
Early
One customer shows a real, large time saving on one deal type; in JLL's survey only 5% of occupiers said they had achieved all their AI program goals.
How it works. AI-native platforms extract terms from leases and offering documents straight into the firm's own Excel model, with every extracted field linked back to the exact document, page and section it came from, cutting a document-heavy CRE workflow from hours to minutes so an analyst reviews and confirms rather than retypes.
Overstated: "AI underwriting" sounds like a machine deciding whether to buy a building. What is live now is document extraction and math; the buy, sell or lend decision still sits with a person.
Shift 6 of 6
Excavators start digging with nobody in the cab
Early
Bedrock said in August that driverless digging was running on three named customer projects, and that month Gravis raised $200 million in what it calls the largest Series A in construction robotics history, to scale autonomous heavy machinery globally, but three projects is a pilot list, not a fleet.
How it works. Bedrock fits excavators with sensors, compute and a machine-learning model; a site manager sets the plan, and the machine perceives the site, plans the motion and digs, stopping automatically if a person gets too close. In August Bedrock announced that its machines were working fully autonomously on three named projects, including sites in Nevada and Texas; at the time these were new machines outfitted at Bedrock's headquarters, and its CTO expected to run on partners' own equipment by the end of the year.
Overstated: 'Fully autonomous' means digging without an operator once a site manager sets the plan, on three named projects; the dump trucks, dozers and loaders Bedrock said in August 'will follow' were a roadmap, not a deployment.
Bedrock Robotics / GlobeNewswire ↗Bedrock Robotics / GlobeNewswire ↗
The scary part
What happens to jobs in Real estate
What AI is already taking:
- Answering leasing calls, texts and emails, booking tours and logging maintenance requests TechFundingNews ↗
- Writing the underwriter's first summary of a mortgage file: sorting the documents, validating the financial data, calculating income and running policy checks PYMNTS ↗
- Extracting lease terms and populating CRE financial models from offering documents Kolena ↗
- Drafting a ready-to-file property tax appeal packet with filing instructions and supporting valuation data HousingWire ↗
What stays human:
Every receipt here stops short of the decision: TD's agent writes the memo and an underwriter decides, Bedrock's excavator digs to a site manager's plan and stops when a person gets close, and EliseAI's Apollo can only do what its user is already allowed to do. Signing for the money, the lease or the crew's safety still needs a named human, and none of these companies claim otherwise.
Get better · 15 minutes
Being fast and skeptical at checking AI's extracted numbers against the source document, the check Kolena built its page-and-section citations around, matters more now than extracting those numbers by hand.
Try this this week:
In Claude, open Customize, then Connectors, and connect Zillow; ask for three homes for sale in a US city you know under a price you pick. Open each result on Zillow.com and compare price and status, then click the tour or agent option and note whose agent and whose lender the next step hands you to: that is the business model of AI home search.
Your move
Leasing and resident AI now has a $4 billion incumbent in EliseAI, and it is moving from answering calls to acting inside the property system, so another leasing chatbot is a crowded bet. The open space is pricing and operations tools that can show a court exactly which data they used: a judge has only temporarily blocked part of New York's law, and the states' RealPage case is going ahead.
Watch for named customers with independently checked numbers: EliseAI's revenue milestone and Dwelly's 300-units-per-agent figure both come from the companies themselves. Red flag in a rent-pricing pitch: any model fed by competing landlords' nonpublic data, the exact practice the states' RealPage case targets.
If your buildings use a revenue-management tool, ask the vendor in writing which inputs come from other landlords' nonpublic data, and keep the answer: a judge has only temporarily blocked part of New York's law, and the states' case against landlords that used RealPage is going ahead. Then test an AI leasing agent on your weakest lead segment, the way Showdigs benchmarked its own tool, not on the channel a human already handles well.
The valuable skill is checking AI output against the source, not producing it by hand: TD's agent writes the underwriter's memo and Kolena links every extracted field to its page, so the person who spots the wrong number fast is worth more. Start by running one lease or loan file you know well through a tool and checking its extraction line by line.
What's actually changing
- 1
Leasing AI graduates from answering the phone to acting inside the property system
How it works. EliseAI's agents started by answering prospects and residents by text, email and phone, booking tours and logging maintenance requests. Its new agent, Apollo, sits inside the platform and inherits each user's existing permissions, so it can act on the property data directly: a leasing agent can ask in plain English for a follow-up to everyone who toured, and the messages go out. Investors then priced the company at $4 billion on $350 million of new money.
Why it matters. An agent that can change records, not just answer calls, is aimed at the leasing office's own workload, and the vendor that holds the system of record collects the budget: EliseAI's pitch is that customers hand it more of their operations every year.
What's overstated'Any task on the platform' is EliseAI's own description of an agent launched in September, and the user evidence at launch was one customer calling it 'the smartest AI I know', not a measured result. It can only do what its user is already allowed to do, so it speeds up a job rather than replacing the judgment behind it.
EliseAIShowdigsVessla
3 receipts
- EliseAI's Apollo agent works inside its platform with the permissions of whoever is using it.
It also inherits each user’s existing permissions, so it sees and does only what that person is already cleared for.
Commercial Observer, 3 Sept 2026 ↗ - EliseAI says its platform powers one in six U.S. apartments.
The platform now powers one in six U.S. apartments, and more than 30 million Americans have interacted with EliseAI since its founding.
EliseAI / GlobeNewswire, 29 Sept 2026 ↗ - Vessla, a new entrant, builds agents that run workflows across existing property software and leave judgment calls to staff.
Its AI agents retrieve information from existing systems, apply company-specific operating rules and execute workflows across connected software, while leaving decisions requiring human judgment or approval to employees.
Vestbee, 19 Oct 2026 ↗
- EliseAI's Apollo agent works inside its platform with the permissions of whoever is using it.
- 2
Rent-setting algorithms move from a software feature to a court question
How it works. States allege that RealPage's software pools competing landlords' nonpublic rent data and recommends prices back to all of them, so rents rise together instead of competing. New York banned using such software to set rents; at the end of September a federal judge temporarily blocked the provision that bars landlords from setting rents on its recommendations, partly because it covers software built on public data as well as nonpublic data. In the same week a North Carolina federal court refused to let RealPage and the landlords that use it out of the states' antitrust case.
Why it matters. Pricing software is the one AI tool in property that lands directly on a renter's bill, so whether its output counts as lawful advice or as price-fixing decides how much of the industry can keep using shared-data pricing.
What's overstatedNeither ruling decides the merits: the judge called it a close question and found RealPage only 'marginally' likely to succeed, and the North Carolina decision only means RealPage and the landlords have to answer the complaint.
RealPageGreystarCamdenPinnacleLivCorWillow Bridge
3 receipts
- California's attorney general says RealPage collects competing landlords' sensitive data and shares it among subscribers.
It does so by amassing competitively sensitive data from competing landlords through its pricing algorithms and sharing this data among subscribers.
California Attorney General, 2 Oct 2026 ↗ - A North Carolina federal court refused to dismiss the states' rent-pricing case against RealPage and kept the landlords in it.
ruling in favor of the coalition, refusing to let landlords out of the case, and denying software company RealPage’s attempt to have the ongoing rent-pricing algorithm case against the company thrown out.
California Attorney General, 2 Oct 2026 ↗ - New York's law covers rent-setting software whether it relies on public or nonpublic data, which supported RealPage's challenge.
The provision applies to covered rent-setting software without distinguishing between systems that rely on publicly available information and those that use nonpublic data, according to MLex’s account of the court order.
PYMNTS, 30 Sept 2026 ↗
- California's attorney general says RealPage collects competing landlords' sensitive data and shares it among subscribers.
- 3
Mortgage files get an AI first reader before the underwriter sees them
How it works. TD's mortgage agent classifies a borrower's documents, extracts and validates the financial data, calculates income, runs policy and consent checks, flags discrepancies and delivers a summary memo before any human reviews the file; PYMNTS reported in May that at TD, pre-adjudication that had averaged roughly 15 hours took under three minutes. Copperlane's agent, Penny, does a similar first pass, scanning bank statements for large deposits that fall outside expected income.
Why it matters. Document review is the slowest, most manual part of getting a mortgage approved; automating the first pass changes how many files one underwriter can carry and how fast a borrower hears back.
What's overstatedTD's agent routes its output to a human underwriter before any decision is made, and in May it covered only pre-adjudication: the rest of the mortgage journey was a plan TD had mapped, not a deployment. A seed-stage 'AI loan officer' is not proof that AI underwrites mortgages end to end.
TDCopperlane
3 receipts
- PYMNTS reports that at TD, pre-adjudication averaged roughly 15 hours before an underwriter could open the file and now takes under three minutes.
At TD Bank, pre-adjudication averaged roughly 15 hours before a human underwriter could open the file. That window is now under three minutes.
PYMNTS, 22 May 2026 ↗ - TD's agent routes its output to a human underwriter before any lending decision is made.
TD’s model includes discrepancy detection as a built-in step and routes output to a human underwriter before any decision is made.
PYMNTS, 22 May 2026 ↗ - Copperlane's AI agent Penny scans bank statements and flags underwriting conditions before a human loan officer reviews the file.
Penny can scan bank statements for large deposits that fall outside expected income, identify other conditions an underwriter is likely to question and proactively contact the borrower for clarification.
HousingWire, 12 June 2026 ↗
- PYMNTS reports that at TD, pre-adjudication averaged roughly 15 hours before an underwriter could open the file and now takes under three minutes.
- 4
Home search moves into chat assistants, on the portals' and the MLSs' terms
How it works. Zillow launched a connector that lets Claude users search homes, see value estimates and check affordability inside the chat, a year after its ChatGPT app. The connector shows Zillow's own listing display and keeps the next step on Zillow: a tour, an agent, a loan officer. Underneath, Cotality's Trestle MCP lets MLS organizations connect approved AI apps to listing data under their existing permissions and licensing rules.
Why it matters. If buyers start a home search inside a general assistant, whoever controls the connector controls the lead, and Zillow built this one so the tour request and the mortgage inquiry still come back to Zillow.
What's overstated'AI home search' here is the portal's own search in a chat window: Zillow says it does not send listing data to Anthropic, and every next step the connector offers, a tour, an agent, financing, runs through Zillow. There are no usage numbers yet.
ZillowAnthropicCotalityRealPlus
3 receipts
- Zillow launched a Claude connector for home search, value estimates and affordability.
Zillow has launched a new connector for Claude, a move that enables consumers to search for homes, review estimated property values and explore affordability within the Anthropic chatbot.
Real Estate News, 20 Oct 2026 ↗ - Zillow says listings keep their broker and MLS attribution and that it does not send listing data to Anthropic.
Zillow said its MLS-approved listing displays retain the broker and MLS attribution used on Zillow.com and that Zillow does not send listing data to Anthropic.
Real Estate News, 20 Oct 2026 ↗ - Cotality introduced Trestle MCP so MLS organizations can connect approved AI apps to listing data under existing controls.
Cotality has introduced Trestle MCP, a no-additional-cost option for MLS organizations on its Trestle platform to connect approved AI applications with authorized listing data while maintaining existing permissions and licensing controls.
Real Estate News, 20 Oct 2026 ↗
- Zillow launched a Claude connector for home search, value estimates and affordability.
- 5
Lease abstraction and CRE underwriting get read by AI first, humans check the flags
How it works. AI-native platforms extract terms from leases and offering documents straight into the firm's own Excel model, with every extracted field linked back to the exact document, page and section it came from, cutting a document-heavy CRE workflow from hours to minutes so an analyst reviews and confirms rather than retypes.
Why it matters. Firms can screen more deals per analyst and turn around a lease review in minutes, but in JLL's survey only 5% of occupiers said they had achieved all their AI program goals.
What's overstated"AI underwriting" sounds like a machine deciding whether to buy a building. What is live now is document extraction and math; the buy, sell or lend decision still sits with a person.
KolenaReal Capital Solutions
3 receipts
- Real Capital Solutions used Kolena's platform on a 150-tenant mall acquisition and cut internal lease review time by 85%.
Internal review time dropped from two hours per lease to roughly seventeen minutes — an 85% reduction
Kolena, 3 Sept 2026 ↗ - 92% of CRE occupiers are piloting AI.
92% of occupiers are also running corporate real estate AI pilots.
JLL / PR Newswire, 28 Oct 2025 ↗ - Only 5% of CRE occupiers report having achieved all their AI program goals.
only 5% report having achieved all their program goals; 47% have achieved two to three goals.
JLL / PR Newswire, 28 Oct 2025 ↗
- Real Capital Solutions used Kolena's platform on a 150-tenant mall acquisition and cut internal lease review time by 85%.
- 6
Excavators start digging with nobody in the cab
How it works. Bedrock fits excavators with sensors, compute and a machine-learning model; a site manager sets the plan, and the machine perceives the site, plans the motion and digs, stopping automatically if a person gets too close. In August Bedrock announced that its machines were working fully autonomously on three named projects, including sites in Nevada and Texas; at the time these were new machines outfitted at Bedrock's headquarters, and its CTO expected to run on partners' own equipment by the end of the year. Gravis raised $200 million to turn existing excavators and other heavy equipment into autonomous machines.
Why it matters. Contractors are short of workers; a retrofit that multiplies what each crew can do changes their costs and schedules, not just their software budget.
What's overstated'Fully autonomous' means digging without an operator once a site manager sets the plan, on three named projects; the dump trucks, dozers and loaders Bedrock said in August 'will follow' were a roadmap, not a deployment.
Bedrock RoboticsGravis RoboticsSundt ConstructionChampion Site PrepZachry Construction
3 receipts
- Bedrock says excavators running its system now operate fully autonomously on live customer sites.
Today, Bedrock Robotics announced that excavators equipped with Bedrock’s system are now operating fully autonomously on live customer sites, shifting AI’s frontier from digital intelligence into real-world production.
Bedrock Robotics / GlobeNewswire, 17 Aug 2026 ↗ - Bedrock's excavators stop automatically when a person or object gets too close.
If a person or unauthorized object gets too close, the excavator stops automatically until the area is clear.
Bedrock Robotics / GlobeNewswire, 17 Aug 2026 ↗ - SoftBank invested $200 million in Gravis Robotics, which turns existing excavators into autonomous machines.
SoftBank has invested $200m in Swiss robotics startup Gravis Robotics, in what has been billed as a record investment.
Tech.eu, 17 Aug 2026 ↗
- Bedrock says excavators running its system now operate fully autonomously on live customer sites.
Who's doing it
14 companies to know, 10 with a round we can source. Newest first where dated.
Search these companies in the startup database ↗ · Explore all research tools
Vesslaproperty operations agents
Stockholm-based agentic operating system that connects the software property owners and managers already use and runs workflows across it, leaving judgment calls to staff.
Latest · 19 Oct 2026Vessla raised a $3.3 million pre-seed round led by Seedcamp to put agents into leasing, building operations and investment planning. Vestbee ↗
$3.3M Pre-seed, 19 Oct 2026 source ↗
EliseAIleasing, resident operations and property-system agents
Agentic AI for multifamily and healthcare operators: answers prospects and residents across voice, SMS, email and chat, and its Apollo agent acts inside the platform with each user's permissions.
Latest · 29 Sept 2026EliseAI raised $350 million at a $4 billion valuation, three months after saying it passed $200 million in annual recurring revenue. EliseAI / GlobeNewswire ↗
$350M Growth round (series not named), 29 Sept 2026 source ↗
KolenaCRE underwriting and lease abstraction
AI platform that extracts data from leases and offering documents into financial models with citations back to the source document.
Latest · 3 Sept 2026Real Capital Solutions deployed Kolena on a 150-tenant mall acquisition and cut internal lease review time by 85%, from about two hours per lease to roughly seventeen minutes. Kolena ↗
Bedrock Roboticsautonomous construction equipment
Fits excavators with sensors, compute and an end-to-end model so they dig without an operator once a site manager sets the plan; in late August its deployments used new machines outfitted at its headquarters, with retrofits of customers' own excavators planned.
Latest · 26 Aug 2026Bedrock's excavators began working fully autonomously on active infrastructure projects in Texas and Nevada. The Robot Report ↗
$270M Series B, 4 Feb 2026 source ↗
Gravis Roboticsautonomous heavy construction equipment
ETH Zurich spin-out whose software turns existing excavators and other heavy machines into autonomous ones.
Latest · 17 Aug 2026SoftBank was the sole investor in a $200 million Series A that Gravis calls the largest in construction robotics. Tech.eu ↗
$200M Series A, 17 Aug 2026 source ↗
DwellyAI-run lettings roll-up
London-based company that buys independent UK lettings agencies and moves them onto its own AI platform for the day-to-day work.
Latest · 28 July 2026Dwelly says each of its agents handles upwards of 300 units, about triple a traditional manager; TNW notes the figures are self-reported. TNW ↗
$170M ($95M equity, $75M debt) Series B, 28 July 2026 source ↗
Copperlanemortgage origination, AI loan officer assistant
AI-native mortgage origination startup whose agent, Penny, reviews borrower documents and flags underwriting conditions for loan officers.
Latest · 12 June 2026Copperlane raised a $4.1M seed round to scale Penny, its AI mortgage loan officer, and did not disclose customer counts or loan volume. HousingWire ↗
$4.1M Seed, 12 June 2026 source ↗
NestoAI-powered digital mortgage lending
Montreal-based digital mortgage lender with tens of billions in originations a year, which also sells its AI-powered lending platform through Nesto Cloud.
Latest · 10 June 2026Nesto raised a CAD $302M Series E at a CAD $1.47B valuation to accelerate its AI-powered lending platform. Nesto ↗
CAD $302M Series E, 10 June 2026 source ↗
Showdigsleasing, self-showing and tour scheduling
AI leasing agent for property managers that responds to leads by text and books self-guided tours against real availability.
Latest · 1 June 2026Showdigs' 2026 platform data shows AI-handled leads booking tours nearly 4 days sooner than leads handled by a human property manager. Showdigs ↗
Xpannerconstruction site automation
Retrofits construction equipment contractors already own with robotics and AI perception, sold as an automation subscription.
Latest · 14 May 2026Xpanner raised an $18M Series B, described as a bridge round, says it reached monthly break-even in 2025 and is on track for full-year profitability, retrofitting equipment customers already own rather than replacing it. Crunchbase News ↗
$18M Series B (bridge), 14 May 2026 source ↗
RealtyAdscommercial real estate leasing marketing and intelligence
AI-native advertising and leasing platform for commercial real estate that now tracks and surfaces broker and prospect engagement in real time.
Latest · 13 May 2026RealtyAds evolved its advertising platform into a Connected Leasing System spanning capture, engage and convert stages of a CRE leasing deal. PR Newswire ↗
Restb.aicomputer vision for property valuation
Image-recognition company whose models read listing photos for property condition and other characteristics, now part of Clear Capital's valuation products.
Latest · 12 May 2026Clear Capital acquired Restb.ai to embed its computer-vision property intelligence into its valuation and floor-plan products. HousingWire ↗
Ownwellproperty tax appeals
AI platform that generates ready-to-file property tax appeal packets for homeowners nationwide.
Latest · 23 Feb 2026Ownwell raised a $50M Series B, $30M of it equity, and launched an AI-generated National Appeals Packet for homeowners outside its main service areas. HousingWire ↗
$50M ($30M equity) Series B, 23 Feb 2026 source ↗
Breezyresidential brokerage, agent workspace
AI operating system for residential real estate agents that consolidates comps, conversation logging, and pipeline tracking into one workspace.
Latest · 2 Feb 2026Breezy launched with an oversubscribed $10M pre-seed round led by Ribbit Capital, opening an exclusive waitlist ahead of a broader U.S. launch. PR Newswire ↗
$10M Pre-seed, 2 Feb 2026 source ↗
Your move
Our read, built on the receipts above.
Leasing and resident AI now has a $4 billion incumbent in EliseAI, and it is moving from answering calls to acting inside the property system, so another leasing chatbot is a crowded bet. The open space is pricing and operations tools that can show a court exactly which data they used: a judge has only temporarily blocked part of New York's law, and the states' RealPage case is going ahead.
Watch for named customers with independently checked numbers: EliseAI's revenue milestone and Dwelly's 300-units-per-agent figure both come from the companies themselves. Red flag in a rent-pricing pitch: any model fed by competing landlords' nonpublic data, the exact practice the states' RealPage case targets.
If your buildings use a revenue-management tool, ask the vendor in writing which inputs come from other landlords' nonpublic data, and keep the answer: a judge has only temporarily blocked part of New York's law, and the states' case against landlords that used RealPage is going ahead. Then test an AI leasing agent on your weakest lead segment, the way Showdigs benchmarked its own tool, not on the channel a human already handles well.
The valuable skill is checking AI output against the source, not producing it by hand: TD's agent writes the underwriter's memo and Kolena links every extracted field to its page, so the person who spots the wrong number fast is worth more. Start by running one lease or loan file you know well through a tool and checking its extraction line by line.
Other industries
- FinanceBanks move agentic AI from chatbots to running back-office and dealmaking work
- Software engineeringSpaceX owns Cursor and starts rewiring it: one Grok bundle, and OpenAI's models on the way out
- HealthcareAmbient scribes go from pilot project to clinical infrastructure
- LegalLegal AI vendors go from research assistant to agent that produces the deliverable
- Sales and marketingCustomers start sending their own AI agents, and vendors race to build the front door for them
- Jobs, HR and recruitingAI runs the first-round interview, not just the resume filter
- InsuranceClaims move from adjuster-first to AI-first, humans supervise the exceptions
- CybersecurityAttackers hand whole intrusion campaigns to AI agents, at about $50 a breach
- Consulting and accountingAI-written client reports become a recurring, public Big Four embarrassment
- Retail and e-commerceAgent checkout retreats to a cart, and the banks guard the last tap
- EducationNYC and LA ban AI for students, not for teachers
- Media and creativeAI video's money goes to the app that resells the models, not the lab that trains them
- Manufacturing and roboticsHumanoid robots move from demo livestreams to paid warehouse shifts
- Transport and robotaxisRobotaxis go from demo to metered fare, and Waymo starts borrowing to scale
- Energy and data-centre powerGrid rules get rewritten mid-buildout, state by state
- GovernmentGSA turns AI adoption into a government-wide subscription, then swaps 'free' for metered contracts
- Agriculture and foodDriverless tractors learn to run the tools behind them, through open platforms and retrofit kits