Episodi

  • AI Agents and Workflow Automation, with Fractional COO Katie Barnes
    Nov 17 2026

    What should you fix before automating a workflow with AI?

    "If we automate it, then we won't have to have that stress there." That is what founders tell Katie Barnes when a handoff keeps breaking. Katie is a fractional COO, an operations chief who works inside several growing companies at once rather than one, and she has spent about 15 years as the second in command to early-stage founders. She tells Chris Grimes and Reuven Gorsht why the work has to be written down before any AI goes on top of it.

    Map the handoff before you automate it

    The handoff Katie hears about most is the one between sales and customer success, the team that looks after a customer once the deal is signed. Before anything is automated, she walks the path a new customer actually takes, talks to the people doing each handoff, writes every step down and tests it until everyone knows who owns what. Only then does the AI get its instructions, and she briefs it the way you would brief an intern: with the full picture, including the odd cases only the glue person, the employee who holds the unwritten knowledge, remembers. If you followed one new customer through your business this week, how many handoffs would you find that nobody has written down?

    Who keeps the AI agents up to date?

    An AI agent is software that carries out tasks for you on its own, and it keeps following the instructions it was given after the business has moved on. Katie expects many companies to discover exploded workflows within the next six months: automations someone set up and walked away from, still running long after everyone forgot them. Her fix is a company brain, a shared wiki the AI reads from, with named owners and a regular rhythm for keeping it current. She also explains why she still holds one-on-ones, even when the AI already has the meeting notes. Who in your company owns the agents once they are running?

    Chapters

    0:00 What a fractional COO does

    2:03 AI agents as a one-person workforce

    2:59 How Katie became a fractional COO

    7:15 Map the process before automating

    11:15 AI tools and shiny object syndrome

    14:42 Handing work to an AI agent

    16:27 Which handoff to automate first?

    18:05 Why AI agents go stale

    22:21 Building a company brain in Notion

    25:41 Do we still need meetings?

    27:28 One-on-ones when AI takes notes

    32:15 AI slop and your own voice

    Your hosts

    Chris Grimes is CEO of FundMore.

    LinkedIn: https://www.linkedin.com/in/chrismgrimes/

    Reuven Gorsht runs Deeded and The Variable.

    LinkedIn: https://www.linkedin.com/in/rgorsht/

    Your guest

    Katie Barnes is an operating advisor and fractional COO at Rise & Optimize, where she helps growing B2B software founders fix operational bottlenecks, clarify who owns what, and run without the founder deciding everything. She writes Systems & Side-eyes, a Substack on building smarter systems and cleaning up operational chaos.

    LinkedIn: https://www.linkedin.com/in/katie-barnes-operations-leader

    Website: https://riseandoptimize.com

    Substack: https://systems.riseandoptimize.com

    Listen and subscribe

    Captivate: https://the-signal.captivate.fm/

    Apple Podcasts: https://podcasts.apple.com/us/podcast/the-signal/id6805026259

    Spotify: https://open.spotify.com/show/59UmJswLjpRO5p7vUwr7hV

    YouTube: https://www.youtube.com/@thesignalreuvenchris

    New episodes every Tuesday.

    Mostra di più Mostra meno
    36 min
  • AI Coordination in Lending: Why Adoption Alone Isn't Enough
    Nov 10 2026

    How should a lender measure AI beyond adoption rates?

    Most companies measure AI by how many licences they bought and how often people log in. Google DeepMind's new essay, The Economics of Intelligence, argues that this misses where the value is: AI pays off less through how well one model performs alone and more through how well people and agents coordinate. Chris Grimes and Reuven Gorsht take that argument into lending, where one loan passes from a loan officer to a processor, an underwriter and funding before it closes.

    Where a mortgage file loses its context

    A mortgage rarely goes wrong inside a single step. It goes wrong in the handoffs, when the loan officer knows the borrower is about to change jobs and nobody writes it down for the underwriter. Chris argues that lenders have chased an automated yes for twenty years and mostly added people, checks and bottlenecks instead, because each tool saw only its own piece of the file. An AI agent plugged into underwriting alone inherits the same blind spot. The hosts also question the human in the loop, the reviewer placed between an AI's output and a final decision, and whether a person who signs off on everything an agent produces is still exercising judgment on a loan. If you traced one file from application to discharge, where would you find the context going missing?

    What a manager is for when agents do the work

    If coordination becomes the bottleneck, the role that has always done the coordinating changes first. Reuven describes management layers as a way to move information up an org chart and standards back down it, and asks whether today's managers will adapt over the next decade or be cut. Chris pictures several AI coordinators working from one shared view of the business, with people kept in the middle for culture and motivation rather than for directing work. Neither host thinks the answer is to buy a suite of AI products and bolt it onto the process you already run. Their test case is a lender whose every rate change sets off about two hundred emails, and the obvious fix is not where the conversation goes. They finish on a scorecard for coordination that starts with how fast a process runs end to end, not how much time each person saves. What would your scorecard measure?

    What was said

    Reuven Gorsht says the DeepMind essay changed his view that AI adoption is mainly a way to raise individual productivity.

    Chris Grimes says lenders have tried to automate the mortgage yes for twenty years and mostly ended up with more people, checks and bottlenecks.

    Reuven says most organizations measure AI adoption by licences and logins, and that he has not seen anyone measure coordination.

    Chris says a single-department pilot will not show the full benefit, because the gains only appear across the whole organization.

    Chapters

    0:00 The economics of intelligence

    1:19 DeepMind's case for AI coordination

    3:46 Where context gets lost

    6:42 Homer Simpson's drinking bird

    9:41 The mortgage automated yes

    12:38 Can one human manage 15 agents?

    13:36 From managers to coordinators

    15:53 AI quarterbacks in leadership

    19:21 Redesigning work at the click level

    22:41 Handoffs in the mortgage process

    29:18 The 200-email rate change

    31:11 A faster horse or a car?

    32:12 Do you need Einstein-level AI?

    34:23 A scorecard for coordination

    36:47 Where to start this weekend

    Mentioned

    The Economics of Intelligence - the Google DeepMind Institute essay by Alex Imas, Andrew Koh and Shane Legg that the conversation starts from - https://institute.deepmind.com/essays/the-economics-of-intelligence/

    The Simpsons - the 1995 episode in which Homer leaves a drinking bird pressing Y on his keyboard at the nuclear plant, Reuven's picture of a human in the loop

    Henry Ford - the adage that customers would have asked for a faster horse, which the hosts use to ask when automating the old process is still the right call

    Your hosts

    Chris Grimes is CEO of FundMore.

    LinkedIn: https://www.linkedin.com/in/chrismgrimes/

    Reuven Gorsht runs Deeded and The Variable.

    LinkedIn: https://www.linkedin.com/in/rgorsht/

    Listen and subscribe

    Captivate: https://the-signal.captivate.fm/

    Apple Podcasts: https://podcasts.apple.com/us/podcast/the-signal/id6805026259

    Spotify: https://open.spotify.com/show/59UmJswLjpRO5p7vUwr7hV

    YouTube: https://www.youtube.com/@thesignalreuvenchris

    New episodes every Tuesday.

    Mostra di più Mostra meno
    38 min
  • Why Banks Blocked Personal AI Agents, and What It Means for Lending
    Nov 3 2026

    Should you connect a personal AI agent to your bank?

    Reuven Gorsht's own research puts about $23 trillion in checking accounts worldwide, earning almost nothing. A personal AI agent, software that acts for you across your email, calendar, messages and accounts, could move that money in minutes. Chris Grimes and Reuven Gorsht have both been running one, OpenAI's Dot, and they ask what happens when it reaches your bank.

    Why the banks said no

    When personal AI agents started touching banking systems, RBC, Bank of America and JPMorgan Chase moved to block them. Security is the obvious reason, but Chris reads it as liquidity, meaning the deposits a bank holds to fund its lending. An agent told to chase the best rate could move a customer's cash whenever another institution paid a fraction of a point more, over a real-time payment rail, a system that settles transfers instantly, like the one Payments Canada and Interac plan to launch in 2027. McKinsey's back-of-envelope estimate puts $170 billion of profit on that idle money. Reuven gives the banks their due: a bank's first job is risk management. If your deposits could move every fifteen minutes, how would you fund your next mortgage?

    Barry, renew my mortgage

    Neobanks, digital banks with no branches, already invite customers to connect Claude or ChatGPT, because they want the market share. Chris expects borrowers to push everyone else the same way. His agent, which he named Barry, already found the errands he had promised to handle across a week of text messages and took care of four of them. Renewing his mortgage is the request he expects next. Reuven notes that about 75% of Canadians simply accept their lender's renewal offer. When a borrower's agent shows up instead of the borrower, what will your renewal process say to it?

    What was said

    Reuven Gorsht says an agent that checks every email and every drive time with you creates decision fatigue of its own.

    Chris Grimes says he set Dot to check in with him on anything external or on a priority account.

    Reuven says he would give an agent a $500 credit card limit for small purchases, but he would not yet connect one to his bank account.

    Chris says he is already testing whether his team's agents can communicate with each other directly.

    Chapters

    0:00 Handing control to AI agents

    0:29 Decision fatigue from AI agents

    3:13 Setting guardrails on Dot

    5:41 What is a personal AI agent?

    7:24 Memory across chat threads

    8:31 An agent that keeps your promises

    11:22 Why banks blocked AI agents

    12:54 Is it really liquidity risk?

    14:42 Friction dollars in checking accounts

    17:56 Real-time rails and the big banks

    20:32 Neobanks take the risk

    21:52 What would make an agent trustworthy?

    24:30 An agent as your day trader

    26:14 When AI writes your emails

    28:23 Personal agents and mortgage lending

    30:51 Barry, renew my mortgage

    33:06 Why Canadians don't switch at renewal

    34:09 Agents talking to agents

    37:07 The 1999 website moment

    38:55 What Chris lets Barry do

    Mentioned

    Dot - the personal AI agent from OpenAI that both hosts are testing, which Chris named Barry - https://openai.com

    RBC - Royal Bank of Canada, named among the banks that moved to block personal AI agents - https://www.rbc.com

    Bank of America - named among the banks that moved to block personal AI agents - https://www.bankofamerica.com

    JPMorgan Chase - named among the banks that moved to block personal AI agents - https://www.jpmorganchase.com

    McKinsey - the source of the back-of-envelope estimate Reuven cites on the profit riding on idle deposits - https://www.mckinsey.com

    Wealthsimple - the Canadian fintech whose bill for moving customers' money in Reuven cites - https://www.wealthsimple.com

    Payments Canada - building Canada's real-time payment rail with Interac - https://www.payments.ca

    Interac - Canada's payments network, named alongside Payments Canada on the real-time rail - https://www.interac.ca

    Your hosts

    Chris Grimes is CEO of FundMore.

    LinkedIn: https://www.linkedin.com/in/chrismgrimes/

    Reuven Gorsht runs Deeded and The Variable.

    LinkedIn: https://www.linkedin.com/in/rgorsht/

    Listen and subscribe

    Captivate: https://the-signal.captivate.fm/

    Apple Podcasts: https://podcasts.apple.com/us/podcast/the-signal/id6805026259

    Spotify: https://open.spotify.com/show/59UmJswLjpRO5p7vUwr7hV

    YouTube: https://www.youtube.com/@thesignalreuvenchris

    New episodes every Tuesday.

    Mostra di più Mostra meno
    39 min
  • Mortgage Operations and AI: Fix the Process First, with Rob Jewett
    Oct 27 2026

    How should a mortgage lender implement AI in operations?

    Before NewFed Mortgage bought AI or hired to meet its volume, its operations team followed loans through the building and counted the clicks. Rob Jewett, Chief Operating Officer of NewFed, an independent mortgage bank based out of Boston, tells Chris Grimes and Reuven Gorsht what he was looking for, and what it took to get the people doing the work to trust what came next.

    Finding capacity before buying it

    Some of NewFed's files were moving forward, then backwards, then forward again, and nobody saw it until someone followed a loan from the very beginning to the very end. Hiring to meet volume works, Rob says, but it is super expensive once training is counted, so he asked a different question first: how much capacity did the team already have that it was not using? That meant measuring how often the same loan was touched. The loop he went after first sits between processing and underwriting, where an underwriter finds a gap and the file goes back to the borrower. NewFed moved that review to the front of the file and worked with Friday Harbor as a design partner, an early customer that helps shape a product before it is finished, so loan officers, processors and underwriters had a say in how it looked and worked. Where would a file in your shop go backwards if you followed it?

    Getting processors and underwriters on board

    Rob does not think employees resist new technology because they dislike change. He thinks they resist it when nobody has explained why they are being asked to change, and he says the fear of being replaced is a reasonable question that deserves a straight answer. His approach is to make the people doing the work the teachers and the software the student. NewFed's AI scenario desk, a tool staff can ask guideline questions instead of searching the guidelines themselves, was trained that way, with staff asking it the same question months apart to see whether the answer held. The hardest group is the veterans with decades on the job, and Rob has a specific answer for them. What would it take for your most experienced underwriter to trust an answer from a machine?

    What was said

    Rob Jewett says a lender should start by identifying where it actually makes its profit, and he judges every decision by cost per loan.

    Rob also argues that roles split across several specialists over the years can start to come back together once bots order credit and disclose loans.

    Chris Grimes says mortgage adopted an assembly line model, one role per step, and that AI means those roles now have to be redesigned.

    Reuven Gorsht says that despite all the investment in technology, the industry has not moved the needle on what it costs to originate a loan.

    Chapters

    0:00 Meet Rob Jewett of NewFed Mortgage

    1:27 From broker shop to operations

    7:21 Why do lenders resist process change?

    9:23 Finding capacity before hiring

    11:00 Why AI rollouts lose buy-in

    16:09 Cost per loan and role redesign

    19:44 Which workflow to fix first?

    22:44 Friday Harbor as a design partner

    26:51 How to choose an AI vendor

    30:18 Getting your people to trust AI

    37:09 Training an AI scenario desk

    41:30 Where should a lender start?

    Mentioned

    NewFed Mortgage - the independent mortgage bank where Rob Jewett is Chief Operating Officer - https://newfed.com

    Friday Harbor - the company NewFed worked with as a design partner on reviewing loan files before underwriting

    The MBA conference - the Mortgage Bankers Association conference, where Reuven and Rob crossed paths

    Your hosts

    Chris Grimes is CEO of FundMore.

    LinkedIn: https://www.linkedin.com/in/chrismgrimes/

    Reuven Gorsht runs Deeded and The Variable.

    LinkedIn: https://www.linkedin.com/in/rgorsht/

    Your guest

    Rob Jewett is Chief Operating Officer at NewFed Mortgage, a retail lender founded in 2001 and licensed in more than twenty states, where he owns the operation end to end. He led the rollout of the company's point-of-sale platform, its e-close and AI tooling, and a data partnership that rebuilt how NewFed reaches people who have already closed with it. HousingWire named him a Tech Trendsetter in 2024 for that work.

    LinkedIn: https://www.linkedin.com/in/rob-jewett77/

    Listen and subscribe

    Captivate: https://the-signal.captivate.fm/

    Apple Podcasts: https://podcasts.apple.com/us/podcast/the-signal/id6805026259

    Spotify: https://open.spotify.com/show/59UmJswLjpRO5p7vUwr7hV

    YouTube: https://www.youtube.com/@thesignalreuvenchris

    New episodes every Tuesday.

    Mostra di più Mostra meno
    45 min
  • Wholesale Lending AI Agents, One at a Time, with Curtis Edwards
    Oct 20 2026
    Should a mortgage lender build its own AI agents?At the TMC conference in Dallas, Reuven Gorsht noticed that the lender sitting next to him had Claude open the whole time and was writing code. That was Curtis Edwards, Chief Lending Officer at Orion Lending, the wholesale mortgage lender he co-founded in 2014, which has built its own technology since the day it opened. He tells Chris Grimes and Reuven what Orion builds, in what order, and why.One AI agent at a timeOrion has decided to build everything itself, but never more than one agent at a time, each with a single job. Its broker portal, in development since the company started, has grown into what Curtis calls basically an LOS, the loan origination system every file runs through. After an insurance agent comes a fraud agent named Sherlock, which follows Fannie Mae's view that fraud is a scheme across many loans rather than one doctored pay stub, so it searches the whole pipeline for the same gift donor, the same person signing verifications of employment, or an employer registered two years ago for a borrower ten years on the job. The end goal is a full underwrite agent, built for consistency, so every broker gets the same transaction from every pod. He puts a number on what it should do for an underwriter's day. Which single job would your first agent do?Who builds when the developers are busy?More than half of Orion's business is now non-QM, loans outside the federal Qualified Mortgage rules that often serve self-employed borrowers, and 40 to 50 percent of those files need an exception to a guideline. Ticket queues in Zoho could not show anyone where a loan stood in that process. Orion's developers had other work, so Curtis built the fix himself: he outlined it in Claude, built the site in Lovable, an app builder that turns a written description into a working web app, and handed it to the developers only to launch behind single sign-on. The Exception Hub went from idea to live in about 30 days. The conversation then turns to the limits of building this way, including an executive's AI assistant on Teams that had to be scaled back. If a leader at your lender built a tool like this next month, who would sign off before it touched a live loan?What was saidCurtis Edwards says Orion will build every agent itself but ship them one at a time, because the alternative is what he calls an AI wrap, a polished interface with nothing underneath it.Curtis also argues that consistency, not speed, is what brokers want from an underwriting agent, so every transaction feels the same whichever team handles it.Chris Grimes argues that because building is now cheap, lenders who spend a downturn building agents will not rehire the way they used to when the market comes back.Reuven Gorsht holds that the friction worth fixing usually sits three or four levels below the leaders who could fix it, which is why leaders need to stay close to the tools.Chapters0:00 Meeting Curtis at TMC Dallas0:47 From car design to wholesale2:34 AEU: hiring AEs out of college7:31 AI-native AEs and call coaching10:12 Purchase lending as rates turn12:49 Insurance and fraud agents16:12 Build vs buy, done responsibly19:05 Sherlock and the underwrite agent21:31 The Exception Hub in Lovable26:00 Prototypes, BAs and agents on Teams28:46 Consumer agents and hands-on leaders30:16 Where is a lender's real friction?MentionedOrion Lending - the wholesale mortgage lender Curtis co-founded in 2014 - https://www.orionlending.comFannie Mae - whose guidance on fraud as a scheme across loans shaped Orion's fraud agent - https://www.fanniemae.comClaude - the AI assistant Curtis used to outline the Exception Hub - https://claude.aiLovable - the app builder Curtis used to build the Exception Hub - https://lovable.devThe University of Texas at Dallas - whose sales program Orion recruits new account executives from - https://www.utdallas.eduTea Talk with Curtis - Orion's podcast on the wholesale mortgage business, hosted by Curtis - https://open.spotify.com/show/5HU7xbZZY07QCzgRVbYklVYour hostsChris Grimes is CEO of FundMore.Reuven Gorsht runs Deeded and The Variable.Your guestCurtis Edwards is Chief Lending Officer at Orion Lending, the wholesale lender he co-founded a decade ago and has helped grow to all 50 states. He leads sales and operations at a lender that builds its own technology, and Orion has been named to HousingWire's Tech100 list two years running. He also hosts Tea Talk with Curtis, Orion's podcast on the wholesale mortgage business.LinkedIn: https://www.linkedin.com/in/curtisedwardsorionlending/Listen and subscribeCaptivate: https://the-signal.captivate.fm/Apple Podcasts: https://podcasts.apple.com/us/podcast/the-signal/id6805026259Spotify: https://open.spotify.com/show/59UmJswLjpRO5p7vUwr7hVYouTube: https://www.youtube.com/@thesignalreuvenchrisNew episodes every Tuesday.
    Mostra di più Mostra meno
    35 min
  • Why Lenders Can’t Take AI’s Word for It
    Oct 13 2026
    Should loan officers build their own CRM with AI?Loan officers building their own customer relationship management systems, or CRMs, with AI was a recurring story at the Digital Mortgage conference in September 2026, celebrated as putting technology in more hands. Sneha Pankaj, Senior Director of Product Management, Mortgage at Cotality, came up through engineering and sees two red flags. She works through them, and the habit that keeps her own assistant useful, with Chris Grimes and Reuven Gorsht.What context as code meansSneha noticed she was re-introducing herself to her AI assistant every day: her role, her product, the problem she was trying to solve. So she stopped. She now commits her call transcripts, demos, pitches and roadmap conversations to a repository, the version-controlled file store software engineers keep their code in, and the assistant draws on all of it from a one-line prompt such as a request for talking points before a customer meeting. She calls this context as code. The idea came from BMAD, the Breakthrough Method for Agile AI-Driven Development, an open-source framework that gives each role in building software, from product manager to architect to tester, its own specialized AI agent and feeds it written context. She says the method has held up for six months and she is piloting it with colleagues in product, marketing and sales. Who checks what the machine did?In mortgage lending, Reuven Gorsht says, one unchecked AI error could cost hundreds of thousands of dollars plus the lender’s reputation. He has seen how quietly it happens. Sneha Pankaj’s answer is evals: a bank of questions with known answers that checks an AI system’s output the way a quality-assurance reviewer would. She says banks already use them. Chris Grimes calls the same need a governance layer, the check on whether an agent is allowed to act at all. He argues the plumbing is arriving before that check, through tools like Salesforce’s headless CRM and MCP servers, the connectors that let AI tools drive business software. That leaves every lender with a practical question: who approves a loan officer’s home-built CRM before it holds customer data?What was saidSneha Pankaj says "loan officers are in the business of building relationships, not building software," and a CRM is a store of personal data that most of its new builders were never taught to protect.Chris Grimes argues that the systems lenders run on were not built for AI, so nothing yet checks whether an agent is allowed to do what it is doing.Reuven Gorsht holds that most people in a lender are operators, not builders, and that the job is finding the few who are.Reuven also holds that you can only catch an assistant's mistake if you have done enough of the work yourself to see the answer is wrong.Chapters0:00 Meeting Sneha at HousingWire1:23 From programmer to product at Cotality4:45 Context as code with BMAD9:31 When Claude has too much context13:32 How much should you trust AI?15:20 Evals and guardrails17:44 The AI governance layer20:03 Loan officers' CRM red flags22:15 Chris's AI-built loan documents23:23 Should everyone be a builder?34:39 Where AI adoption stands37:54 AI five years from nowMentionedCotality - property data and analytics company, formerly CoreLogic - https://www.cotality.comBMAD - open-source framework that gives each role in software development its own AI agent - https://github.com/bmad-code-org/BMAD-METHODClaude - AI assistant that can work across a folder of files - https://claude.aiVisual Studio Code - widely used code editor that can host AI assistants - https://code.visualstudio.comSalesforce - CRM platform with a headless CRM and an MCP server for connecting AI tools - https://www.salesforce.comYour hostsChris Grimes is CEO of FundMore.Reuven Gorsht runs Deeded and The Variable.Your guestSneha Pankaj is Senior Director, Mortgage Data Solutions at Cotality, formerly CoreLogic — the #1 provider of property data and analytics for real estate, mortgage, and insurance. She spent her early tenure at Cotality leading engineering before moving into product leadership, and today she’s taking a 0-to-1 mortgage retention product from business case to commercialization, in a market where lenders lose up to 80% of their book of business after closing. She’s also been a driving force behind AI adoption at Cotality — leading company-wide AI experimentation efforts and pushing AI fluency beyond engineering into product and sales. She brings a systems thinker’s discipline to product: roadmap thinking, client context, and decisions all committed to the same repository as the code.LinkedIn: https://www.linkedin.com/in/snehapankaj/Listen and subscribeCaptivate: https://the-signal.captivate.fm/Apple Podcasts: https://podcasts.apple.com/us/podcast/the-signal/id6805026259Spotify: https://open.spotify.com/show/59UmJswLjpRO5p7vUwr7hVYouTube: https://www.youtube.com/@thesignalreuvenchrisNew episodes every Tuesday.
    Mostra di più Mostra meno
    41 min
  • AI Finance: The Last-Mile Transformation and ROI
    Oct 6 2026
    Applying for a mortgage inside ChatGPT: what lenders loseChris Grimes went looking for a raincoat before a trip to Scotland, typed what he wanted into ChatGPT, and bought one without ever opening a shop's website. Applying for a mortgage inside ChatGPT is the same move with three more zeroes on it, and it is already live. Better now takes a mortgage application without the borrower leaving the chat window, and at the other end of the same business the model companies are pushing into the parts of lending that were supposed to be too complex and too regulated to touch.WHAT CHANGED WHEN THE MORTGAGE APPLICATION MOVED INSIDE CHATGPTThe enterprise end moved first, and quietly. Google announced a partnership with a large financial institution, Microsoft has done the same, and Anthropic is doing it too. Reuven Gorsht's reading of why is unsentimental. Companies valued in the trillions after four or five years have to show a buyer that a dollar in returns three or five or ten, and the strength of the core model alone no longer does that, because most work gets done perfectly well on the cheaper models. So the labs are hiring the people who already have the relationships. Reuven has watched executives from SAP, Oracle and Salesforce move across to companies like Anthropic, arriving with a Rolodex and a familiar face at the office of the CIO.The consumer end moved more loudly. Better's application runs inside ChatGPT, so a borrower starting a mortgage there never opens the lender's own interface. Chris expects a dedicated consumer finance surface to follow, with a connector like Plaid pulling the full picture together so the assistant can give advice against real balances rather than guesses. His worry is not that the labs want to own the application layer. It is that they are dissolving it, folding what used to be somebody's product into something the assistant simply knows how to do.WHY THE ADVICE LAYER IS THE EXPOSED ONEReuven's argument is that consumers take the path of least resistance every time, and that the evidence is already in. He watches Amazon vans on his street while the stores in his local plaza close. He bought his wife a Tesla, a purchase in the mid sixty thousands, entirely through an app, and has never met anybody from the company. Keynote speakers at the New England Mortgage Bankers Association put the same point to the industry more bluntly, which is that the trade says borrowers want a relationship when what they actually want is the lowest rate.Chris has held a version of this since before he left brokering. His view is that the mortgage was never the thing anybody was buying. The house is. The mortgage is the monthly payment you stop thinking about the week after it funds, which makes the advice wrapped around it, rather than the loan itself, the part that is genuinely exposed. Neither host believes people vanish from a purchase this stressful. Both of them get stuck on the same question, which is where exactly those people are supposed to stand once the friction is gone.THE GAP BETWEEN AI SPEND AND AI RETURNThe last stretch of the episode is about money, and it is where both hosts turn sceptical. Reuven cites a presentation from one of the world's largest asset managers on the scale of capital expenditure going into AI infrastructure this year, the gigawatts of data centre capacity still to be built, and what a single investor's portfolio companies are spending with Anthropic annually. The figures are striking, and they arrive with a caveat the episode states out loud, which is that the firm presenting them has a great deal of its own capital deployed in exactly what it is describing.Chris's question is what happens when spending on that scale meets an antitrust regime and a compute shortage at the same time. Reuven's question is simpler and more uncomfortable. He asks what anybody is actually getting back, puts a figure on how much of his own AI output he never uses, and extends it to companies doing the same thing with far larger budgets. WHAT YOU'LL TAKE AWAY- The model companies are entering lending from the borrower's end and the capital markets end at once, which puts the pressure on the middle of the stack rather than on any single product.- What a borrower is buying is the house, so the advice wrapped around the mortgage is the layer most exposed when the friction disappears.- Price and friction are the two things either host says a consumer genuinely chooses on, and which of the two a lender is worse at.- Money going into AI is not the same as value coming out, and the honest version of that gap is easier to hear from somebody describing their own usage than from any vendor.CHAPTERS(00:00) Every Major AI Lab Has Launched a Finance Tool(00:59) Why the Model Companies Need an ROI Story(03:00) Enterprise Partnerships and the App Layer(07:53) What Your LLM Already Knows About You(10:16) Buying Through the Chat Window(14:05) What Borrowers Say They Want Versus What They Pick(16:59)...
    Mostra di più Mostra meno
    37 min
  • The Moat Is Not the Software
    Sep 29 2026
    Scott Larter runs BrokerBot on licensed Canadian property data, and he came to mortgage from the data side, at Opta, Verisk and FCT. Forty-four minutes in, he puts a three to five year clock on his own company's defensibility and names what he thinks is actually holding the door shut. Chris Grimes and Reuven Gorsht push on what is left to defend once a competitive analysis that used to take a team weeks takes twenty minutes and a coffee, and once the same intelligence sits in everybody's pocket. What the three of them land on is narrower than what the industry has been selling, and none of it is the software.Why AI is not a competitive advantage in mortgage techIn 2018 Scott Larter put a big orange booth on a trade show floor in Montreal, knew one person in the room, and was there to find out whether property data had any place in mortgage lending at all. Eight years later he runs BrokerBot, and he is arguing about a different question. What counts as a competitive advantage in mortgage tech now that the cost of building the product has fallen to almost nothing?Scott came to mortgage from the data side, at Opta, Verisk and FCT, which is why he treats a broker's client book as a dataset in an industry that has always treated it as a rolodex. He is talking to Chris Grimes and Reuven Gorsht, who between them have built a lending decisioning platform and a real estate closing business. This page sets out what the three of them worked through and which questions they left open. What they conclude is on the tape.WHY THERE ARE SUDDENLY HUNDREDS OF VENDORSReuven's framing is that the economics of starting a software company have inverted. When he started MoveSnap and when Chris started FundMore, you validated an idea hard first, because you were about to spend at least a million dollars in cash or in sweat equity. Now anybody can build a working prototype in an afternoon, so the constraint has moved.He brings a figure back from a panel of venture investors he sat in on a few weeks before recording. Their position was that they are still funding roughly the same number of deals a year that they always have, while the denominator underneath that number has grown by something like a hundredfold. Scott's experience points the same way from the other side of the table. In his data years he was approached by ten to twenty startups a year looking for Canadian property data to build on, and he is blunt that cheap Canadian property data does not exist in this country.Chris supplies the uncomfortable part. He argues that low adoption among lenders is not a temporary lag, that it has stayed low, and that a proof of concept which never reaches production is not evidence of anything. WHAT COUNTS AS A COMPETITIVE ADVANTAGE IN MORTGAGE TECH NOWReuven's answer begins by taking something away, in one flat sentence that is the most quotable line on the tape. What he says is left afterwards sits in two places. The first is distribution, meaning a network or a community you can actually put a product in front of. The second is the judgment to know which output is any good, which he ties to twenty-five years of knowing which advertisement is off message and which one will get you in trouble with a regulator.Scott gives the clearest worked example. He ran a competitive analysis across the Canadian and US markets and turned it into a deck for his team, and he puts a figure on how long that took against the weeks it used to take. A moat is not a moat if it can be replicated, and that today almost everything can be. You can point a model at a competitor and have it decompose what they are doing in fifteen minutes.He also raises the problem nobody in the industry enjoys looking at. When you feed your own working into a general purpose tool, you may be sharpening the thing a competitor uses against you next quarter. He puts it as a question, because he does not have the answer either.THE END OF THE PURE PRODUCT COMPANYScott asks the hosts a direct question about their own businesses. If a junior person with some AI ability can now build internally, for a salary, what a vendor charges in licensing, does that change how a lender evaluates products like theirs?Chris gives ground immediately and says that it will. Then he draws a line, and where he draws it is the argument of this section. He does not think vendors get wiped out. He thinks the pure product company stops existing unless it owns one very specific problem in a very large space, and that what replaces it is a partner who brings most of the way and then works with the customer on the rest. He puts the split in proportions and calls the result a pivot toward services. Whether he is right is the thing to listen for.Reuven works the same idea through a different industry. He describes replacing an advertising agency brief with a Sunday morning and a model, and is precise about what he did not get, which was thirty years of agency expertise and somebody else carrying the ...
    Mostra di più Mostra meno
    44 min