Darragh Buckley on Increase's bank charter, AI in fintech, and the 'no abstractions' engineering philosophy
Summary
Darragh Buckley discusses Increase's transition to a bank charter to gain direct Federal Reserve access and avoid third-party banking provider complexities. He emphasizes a 'no abstractions' API philosophy that exposes underlying financial infrastructure details to engineers. Buckley warns against outsourcing accountability to AI, advocating for human sense-checking and citing Chesterson's fence to respect regulatory constraints. The conversation also covers the potential for agentic commerce to support local businesses and the need for isolated payment instruments for AI agents.
Episode sections
- 0:23 – 5:03
Early Stripe Culture and Urgency
Darragh Buckley recounts joining Stripe in 2010 and describes the company's culture of shared urgency, illustrated by an anecdote about founders John and Patrick running from an illegal parking spot. He emphasizes that this urgency was a positive, energizing force rather than a top-down mandate.
- 5:03 – 12:01
Written Communication and AI in the Workplace
Buckley explains how Increase maintains Stripe's practice of long-form written communication to ensure clarity and accountability. He discusses the 'don't expect me to read it unless you read it' principle in the context of AI-generated content, noting that AI is best used as a thinking partner for the user rather than a replacement for human responsibility.
- 12:01 – 19:03
Scaling and the 'Worry Budget'
The speakers discuss the trade-off between hiring entry-level employees for training versus using AI for immediate high-competency output. Buckley states that Increase remains conservative with headcount due to its banking charter but refuses to outsource its 'worry budget'—the sense-checking and accountability functions—to AI agents.
- 19:03 – 22:00
Increase's 'No Abstractions' Philosophy
Buckley details Increase's API design principle of 'no abstractions,' contrasting it with Stripe's approach. Increase aims to pass through underlying financial network details (like FedACH codes) with high fidelity to engineers, allowing them to build directly on the metal of the banking infrastructure rather than hiding complexity.
- 22:00 – 28:07
Engineer-Led Decision Making and Compliance Risks
The conversation shifts to the rise of engineer-led decision-making in fintech. Buckley warns that while empowering engineers is valuable, it carries risks if they ignore regulatory constraints or dismiss institutional experience, citing 'Chesterson's fence' as a reason to respect existing rules.
- 28:07 – 30:05
Navigating Regulatory Barriers and Internal Culture
Speaker_0 discusses the semi-permeable nature of financial regulations, distinguishing between physically possible but legally prohibited actions and those that are merely difficult. Speaker_1 advises against internal antagonism between engineering and compliance, advocating for a shared struggle against problems rather than adversarial relationships.
- 30:05 – 32:25
AI's Impact on Organizational Dynamics
Speaker_1 suggests that AI will reduce the need for 'warm bodies' in routine roles, allowing humans to focus on expertise. Speaker_0 agrees, predicting a future where employees respect human colleagues as experts rather than viewing them as obstacles, reserving friction for interactions with AI systems.
- 32:25 – 37:32
Increase's Transition to a Bank Charter
Speaker_1 confirms that Increase purchased a bank approximately 1.5 years prior to the interview and rebranded it as Increase Bank. He explains that this move allows for direct integration with the Federal Reserve and greater control over product experience, avoiding the technical debt and misalignment associated with third-party banking-as-a-service providers.
- 37:32 – 40:41
Product Velocity and Marketing Leverage
Speaker_1 highlights that owning a charter enables faster product iteration, such as implementing FedNow, without needing to justify technology spend via traditional cost-benefit spreadsheets. He argues this serves as a marketing signal to ambitious tech companies, demonstrating that Increase is aligned with their roadmap and needs.
- 40:41 – 45:37
Agentic Commerce and Small Businesses
Speaker_1 expresses bullishness on agentic commerce, particularly its potential to help small and local businesses by allowing AI agents to order from local merchants when consumers are time-deficient. Speaker_0 adds that AI can improve matching for long-tail, hyper-specific products, potentially increasing willingness to pay for unique, high-quality goods.
- 45:37 – 48:22
Banking Needs for AI Agents
Speaker_1 identifies isolated accounts and payment instruments as critical needs for AI agents, arguing that convenience, rather than increased trust, will drive adoption. Speaker_0 contrasts the 'Amazon model' (centralized) with the 'Shopify model' (merchant-aligned), suggesting banks should focus on the latter to support diverse agent use cases.
Claims worth verifying
Darragh Buckley joined Stripe in 2010 in Palo Alto with founders Patrick and John.
Unidentified speaker (speaker_1) ·1:34
How to verify: Darragh Buckley's LinkedIn profile and Stripe's historical records or early employee lists. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Greg Brockman worked at Stripe in its early days.
Unidentified speaker (UNKNOWN) ·1:53
How to verify: Greg Brockman's LinkedIn profile or Stripe's early team announcements. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Increase processes more than $500 billion annualized for customers like Coast, Gusto, and Ramp.
Unidentified speaker (speaker_0) ·0:02
How to verify: Increase's official website or press releases verifying their AUM/TPV figures. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Increase's API design principle is 'no abstractions,' passing through underlying network details like FedACH codes with high fidelity.
Unidentified speaker (speaker_1) ·20:05
How to verify: Increase's developer documentation or blog post on their API design philosophy. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Primary sourceStripe's early product approach was to create a 'beautiful abstraction' over payments.
Unidentified speaker (speaker_1) ·20:05
How to verify: Stripe's historical API documentation or public statements by Patrick and John Collison. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Increase is a bank with a charter, allowing it to build directly on the Federal Reserve network.
Unidentified speaker (speaker_1) ·13:45
How to verify: FDIC or Federal Reserve membership lists for Increase. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Chesterson's fence is a principle for respecting regulatory constraints without understanding their origin.
Unidentified speaker (UNKNOWN) ·26:38
How to verify: General knowledge of the Chesterton's Fence philosophical principle applied to regulatory compliance. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Increase purchased a bank approximately 1.5 years before the interview and rebranded it to Increase Bank.
Unidentified speaker (speaker_1) ·33:18
How to verify: Check Increase Bank's corporate history or press releases from mid-2026 for the acquisition announcement and rebranding. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Stripe was built on top of Wells Fargo.
Unidentified speaker (speaker_0) ·35:23
How to verify: Review historical documentation or interviews with Stripe founders regarding their initial banking partner. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Increase technology processes more than $500 billion annualized in payments, card transactions, and loans.
Unidentified speaker (speaker_0) ·39:04
How to verify: Check Increase's official website or press releases for updated AUM (Assets Under Management) or transaction volume metrics. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Companies like Coast, Gusto, and Ramp rely on Increase technology.
Unidentified speaker (speaker_0) ·39:04
How to verify: Check Increase's customer case studies or press releases for partnerships with these specific companies. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Speaker_1 was Stripe's first employee.
Unidentified speaker (speaker_0) ·39:24
How to verify: Check Darragh Buckley's LinkedIn profile or Stripe's historical employee records. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Increase purchased a bank charter approximately 1.5 years prior to the interview.
Unidentified speaker (speaker_1) ·33:18
How to verify: Search SEC filings or press releases from Increase/Bank of Increase for the acquisition date. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Primary sourceFedNow requests for payments are difficult to justify via traditional cost-benefit spreadsheets.
Unidentified speaker (speaker_1) ·37:55
How to verify: This is a subjective claim about internal banking economics; verify by reviewing industry analyses on FedNow adoption costs and ROI challenges for non-chartered entities. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Speaker_0 has a five-year-old daughter who lost a headband at a soccer game.
Unidentified speaker (speaker_1) ·45:17
How to verify: This is a personal anecdote; verify via social media posts or personal interviews with Speaker_0 if available. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Speaker_0 wants a 36-inch long felling axe forged in Sweden by specific blacksmiths.
Unidentified speaker (speaker_0) ·43:37
How to verify: This is a personal preference; verify via social media posts or personal interviews with Speaker_0 if available. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Increase Bank is aligned with ambitious technology companies.
Unidentified speaker (speaker_1) ·36:33
How to verify: Check Increase Bank's marketing materials, website, and product documentation for their target customer segmentation. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.
Do your own research
- Verify Increase Bank's FDIC insurance status and charter details via the FDIC database or OCC public records.
- Check Increase's API documentation for 'no abstractions' claims and compare technical specifications with Stripe's API to assess the difference in data fidelity.
- Review Increase's public filings or press releases regarding the acquisition of the bank charter to confirm the timeline and regulatory approvals.
- Research 'Chesterson's fence' principles in fintech compliance to understand the historical context of wire transfer reporting rules.
- Analyze the market adoption rates of FedNow and other real-time payment rails to assess the competitive advantage of direct Federal Reserve access.
- Investigate the regulatory landscape for AI agents handling financial transactions, including current guidelines from the CFPB or Federal Reserve.
- Look for case studies or pilot programs involving agentic commerce for small businesses to evaluate the viability of the 'local commerce' thesis.
- Check for any conflicts of interest or disclosures from Darragh Buckley or Increase regarding their AI tooling partnerships or investments.
Full transcriptexpand
- 0:02
- Unidentified speaker (speaker_0): This episode is brought to you by Increase. Increase offers API first banking for ambitious technology companies. Coast, Gusto, and Ramp rely on Increase to process more than $500 billion annualized. If you're building the next great financial product and need a bank that gets it, visit
- 0:21
- Unidentified speaker (speaker_0): Increase.com.
- 0:23
- Unidentified speaker (speaker_0): Dara, thanks so much for coming back on the podcast. Cheers, thank you for having me. Oh, of course. It's my pleasure. Um, I am very much looking forward to this conversation. I have a uh simple three-act structure for us today. So uh we're gonna we're gonna talk a little bit uh company building and culture stuff, which are selfishly just questions I've always wanted to
- 0:43
- Unidentified speaker (speaker_0): ask you. And I figured what better place than a podcast? We're gonna going to talk about getting close to the metal and sort of building into in financial services and some kind of deep infrastructure stuff. And then, of course, as we are legally obligated to do, we will talk AI, but we'll make that rapid fire and uh we'll do it at the end. So to
- 1:03
- Unidentified speaker (speaker_0): start with, you before your current role, uh, way back in the day, were the first employee at Stripe. And to be honest with you, I don't really have a question here. I was just hoping that you could tell us.
- 1:15
- Unidentified speaker (speaker_0): one or two interesting stories maybe from the early days. Uh you know, ideating uh with John and Patrick before the company started, challenges and breakthroughs early on, but just uh give us one or two war stories to start us off.
- 1:30
- Unidentified speaker (speaker_1): Um well, hopefully more happy than war stories. Um
- 1:33
- Unidentified speaker (UNKNOWN): so yes,
- 1:34
- Unidentified speaker (speaker_1): I uh um I joined up with uh with Patrick and John. in 2010 down in Palo Alto for the company that grew into uh grew into Stripe. Um I think, you know, since then, you know, came to realize how fortunate a uh a group that I got to work with, right? Like uh
- 1:53
- Unidentified speaker (UNKNOWN): early
- 1:53
- Unidentified speaker (speaker_1): days Stripe, you get to kind of sit down and you're working through, I don't know, payment card industry data security standards are on the table. And there's Patrick, there's John, there's Greg Brock. Brockman. And you know, there I am. And it's on the reflection, it's uh it's a fun group to be working with. Um uh it was my introduction to technology startups. Uh uh
- 2:13
- Unidentified speaker (speaker_1): so getting to see kind of uh a growth curve like Stripe is just fun and fascinating. Um
- 2:19
- Unidentified speaker (UNKNOWN): people
- 2:20
- Unidentified speaker (speaker_1): often ask what kind of parts of early Stripe could be ported, right? And can be uh uh kind of you know instilled within it. a different organization. And uh if you'll forgive me a bit of a story here, I'll I'll tell one. There's something very motivating in John and Patrick's actions. And like I it I think it's purposefully practiced. And so the story
- 2:39
- Unidentified speaker (speaker_1): goes that uh we're just moved into this nice little, you know, cute little cubby office in in Palo Alto per tradition. You need to kit out the the office with, you know, six tables and changes. chairs and um what do you do but go find the cheapest ones that you possibly can? And that's a a trip to IKEA. Car arrives back and it's in the driveway
- 2:59
- Unidentified speaker (speaker_1): after being loaded up. And you go out and you grab a box and you kinda trudge it in and you yeah you place it down. And uh I look around and John and Patrick
- 3:08
- Unidentified speaker (UNKNOWN): are
- 3:08
- Unidentified speaker (speaker_1): like sprinting Printing the other direction. Uh and there's do we park illegally? Did you know uh
- 3:15
- Unidentified speaker (speaker_1): is this like not a nice part of Palo Alto? Right.
- 3:18
- Unidentified speaker (UNKNOWN): Like,
- 3:19
- Unidentified speaker (speaker_1): what's up? Why why are we running? Yeah.
- 3:20
- Unidentified speaker (UNKNOWN): And
- 3:21
- Unidentified speaker (speaker_1): so, you know, you see other people running. It's socially unacceptable not to run yourself.
- 3:25
- Unidentified speaker (UNKNOWN): So yeah, sure. Yeah,
- 3:26
- Unidentified speaker (speaker_1): yeah, up and run too. And um, it turns out that this is just how they do things, right? Like there's
- 3:31
- Unidentified speaker (UNKNOWN): just like an instilled
- 3:32
- Unidentified speaker (speaker_1): sense of urgency that happens uh in just about everything. And not in a like obnoxious way, not in a
- 3:37
- Unidentified speaker (UNKNOWN): like,
- 3:37
- Unidentified speaker (speaker_1): why haven't you done it fast? faster? Right. Like
- 3:39
- Unidentified speaker (UNKNOWN): not in
- 3:40
- Unidentified speaker (speaker_1): a like kind of browbeating, top-down dictatorial uh fashion, but in like, oh, we're all in this together, and of course we're going to just do it at, you know, the fastest way that can we can do it. And like,
- 3:50
- Unidentified speaker (UNKNOWN): why wouldn't
- 3:50
- Unidentified speaker (speaker_1): we do it just uh why wouldn't we like be sweating and uh running to do this? It's fun, right? Like it's it's a very energizing group to be part of. And I feel like just more things could do with just that urgency for urgency. urgency's sake, and not in a kind of in a shared sense of struggle, right? Like just in a like, of course we're not going to walk to do this. Of course we're going to
- 4:10
- Unidentified speaker (speaker_1): like bring energy to this domain. I know it's it's a great one that you can kind of port everywhere, whether you're want to, you know, you're moving boxes in a house or whether or not you're building a company or whether or not you're helping your kids do homework. You you can you can just instill urgency uh in a rewarding way.
- 4:25
- Unidentified speaker (UNKNOWN): That's
- 4:25
- Unidentified speaker (speaker_0): a good one. Yeah. I mean I it's it's funny. It's like that that intrinsic urgency because there is a lot of, I think, like performance urgency uh in texts and in in startup land. Uh and so there is very much, I think, a difference between those two things.
- 4:43
- Unidentified speaker (UNKNOWN): And
- 4:43
- Unidentified speaker (speaker_0): I I guess I'm curious, how have you applied that or other kind of lessons you took from those days to increase? Because I mean, one of the things that I'm I'm constantly struck by with Stripe and with kind of the DNA of Stripe as it's sort of spread out across the industry is it does seem very intentional about
- 5:03
- Unidentified speaker (speaker_0): culture and about sort of some of these things that are can can become like soft, squishy buzzwords, uh, unless you think very intentionally about it. What have you applied at Increase?
- 5:15
- Unidentified speaker (speaker_1): Urgency is certainly one, right? Uh and trying to convey that and convey it in the same kind of, you know, spirit of shared enterprise way. There are some practices as well, which are uh very useful to port. And I think these are like relatively prevalent now within uh the technology industry. But things like long form written as your standard way of like uh first of all
- 5:35
- Unidentified speaker (speaker_1): thinking, but also communicating your decision. Sure.
- 5:37
- Unidentified speaker (UNKNOWN): Right.
- 5:37
- Unidentified speaker (speaker_1): And uh being able to inspect your own thoughts, uh uh both, you know, in present moment, but also when future you looks back. Um uh I wonder why I thought that crazy thing was worthwhile. It's useful to be able to look straight to the source. It's useful for gathering input from others, it's useful for uh, you know, having the habit, for example, of attaching
- 5:57
- Unidentified speaker (speaker_1): citations when you're uh loosely stating we hear about this many times, and if you have the forced habit of needing to, you know, cite the sales notes or the support request for it, uh you may suddenly realize that you haven't heard about it many times. The long form written habit, I think uh um it certainly
- 6:17
- Unidentified speaker (speaker_1): existed before Stripe, but Stripe placed a different emphasis on it than others and increase maintains and uh uh emphasizes that too.
- 6:23
- Unidentified speaker (speaker_0): Yeah. Well, I mean, I I wanted to ask you about that. specifically because one of the other things, and I'm starting to see a little bit more awareness of this or a little bit more sort of self-reflection about this, is how a lot of these practices and quite frankly, a lot of these companies that are being built sort of
- 6:43
- Unidentified speaker (speaker_0): do so in the age of AI. So this will be like an AI bonus question that I'll ask now before we get to the real AI uh question round at the end. But I I am curious because like there was a something going around on Twitter. I can't remember what the which company it was, but they had written down their policies for how they expect their employees to write or not
- 7:03
- Unidentified speaker (speaker_0): write with AI, right? And it was sort of stating these things that we all know are true, but that were it felt like for a while we were pretending like weren't true. Things like, well, to a certain degree, AI does. replace your thinking. Or uh if you use it in a bad way, uh, you know, AI will actually create more work for the person
- 7:23
- Unidentified speaker (speaker_0): reading it than it was work to put the thing together in the first place. Or, you know, if you use AI, it really does sort of risk, you know, loose citations or misplaced citations or missing citations, things like that. And yet at the same time, having said all of that, and I'm a writer, so I have sort of my own proprietary feelings about the use of
- 7:43
- Unidentified speaker (speaker_0): AI. But even beyond that, it does feel like we've sort of entered a different era in terms of like scaling companies, thinking about units of intelligence versus units of human labor. Like there's just a lot of things that uh feel like they've sort of fundamentally shifted and that we won't necessarily be
- 8:03
- Unidentified speaker (speaker_0): going backwards on at any point in the future. I mean, you guys have a small team at Increase. How do you think about like scaling and about retaining the importance of things like a culture of writing as an example? Now that we have these tools.
- 8:18
- Unidentified speaker (speaker_1): Yeah, I think um, like societally, we're getting better with some parts of this. Uh, in particular, um, like uh don't expect me. as a receiver to read something if you as the, you know, um titular writer did not, in fact, write it, right? Uh
- 8:33
- Unidentified speaker (UNKNOWN): the slab
- 8:34
- Unidentified speaker (speaker_0): grenade concept. Yeah,
- 8:35
- Unidentified speaker (UNKNOWN): yeah.
- 8:36
- Unidentified speaker (speaker_1): Yeah. The the um um you surely must have read this more times than than than I would if uh what you're looking for is to to communicate this. And some of this can be obvious, right? Like, oh, in an email or like you
- 8:47
- Unidentified speaker (UNKNOWN): know,
- 8:47
- Unidentified speaker (speaker_1): in a slack message or in a long form written document. Some of it is less. obvious though, right? Like as a like pull request description, right? The robots will helpfully write you a pull request description that may or may not be like apt by the time the pull request is merged. And if, you know, that was our prior way of communicating, you know, the kind of the
- 9:07
- Unidentified speaker (speaker_1): intention of the code. And if the robot writes it and things have changed, you're you're, you know, what am I doing reading it? Um
- 9:15
- Unidentified speaker (speaker_1): and so as uh we're still seeing some the principle of of don't expect me to read it unless you read it. Kind of I will um I will value your time um uh as much, if not more, than mine. And we're seeing kind of like the edges of where that ends up bleeding through. Uh those are important using robots as thinking tool, too, right? Like I'm
- 9:34
- Unidentified speaker (UNKNOWN): not sure have
- 9:35
- Unidentified speaker (speaker_1): you had this like joyful experience? Uh oh, I'm going to say something now which I might regret on a podcast, but like let me go ahead and say it. Uh yeah.
- 9:43
- Unidentified speaker (UNKNOWN): Uh
- 9:43
- Unidentified speaker (speaker_1): which is if you've ever uh you know been in a long car ride, and in addition, or as a trade to listening to a podcast, you you're you're like using your favorite robot on its favorite audio mode to have an interactive conversation about a topic that you almost understand. It's like
- 9:59
- Unidentified speaker (UNKNOWN): most uh like
- 9:59
- Unidentified speaker (speaker_1): best when you almost understand it. Like uh for real. what's the like premium structure in a small business administration lending program?
- 10:07
- Unidentified speaker (UNKNOWN): Right. Like, you
- 10:07
- Unidentified speaker (speaker_1): know, and like
- 10:08
- Unidentified speaker (UNKNOWN): that
- 10:08
- Unidentified speaker (speaker_1): doesn't make sense because like that will be contrary to this thing. And I love that mode of thinking interaction, right? Like
- 10:14
- Unidentified speaker (UNKNOWN): I love
- 10:14
- Unidentified speaker (speaker_1): that mode of of of learning and particularly useful features there, like I want to be able to interrupt you. Like, no, no, no, I know the first pass version, right? Like skip ahead. Uh right,
- 10:24
- Unidentified speaker (UNKNOWN): right, right.
- 10:24
- Unidentified speaker (speaker_1): For real, how is that com uh uh consistent with this? other thing? And like you're allowed to admit it's not. It my particular like learning moments that I've enjoyed with uh with the robots have certainly been like not as like generating this content for someone else, but generating it for myself.
- 10:39
- Unidentified speaker (UNKNOWN): Yeah,
- 10:39
- Unidentified speaker (speaker_0): I mean kind of using it to to push back or to argue something out. I mean, I I actually have used it more recently like while walking around, you know, and so you just sort of turn on like the voice mode and you just sort of like like argue with it, you know. And it's fun. The other thing I like about it is that you can interrupt it as much as you want and it stays endlessly patient with you. Like it doesn't get mad if
- 10:59
- Unidentified speaker (speaker_0): you just interrupt it 9,000 times. Uh and it's like, oh no, great question. I'm sorry I skipped over that, blah, blah, blah. And it's just like endlessly, I mean you can kind of see the argument for it in an education context if it's appropriately deployed, because it's
- 11:12
- Unidentified speaker (UNKNOWN): just like
- 11:12
- Unidentified speaker (speaker_0): endlessly patient, you know.
- 11:14
- Unidentified speaker (UNKNOWN): If we
- 11:15
- Unidentified speaker (speaker_1): learn how to use it in an education context, we aren't there, certainly. No,
- 11:18
- Unidentified speaker (UNKNOWN): no, no.
- 11:18
- Unidentified speaker (speaker_1): But like uh if we learn how to, it'll be fantastic. It's amazing how much interruption is needed for a good conversation, right? Like it's
- 11:25
- Unidentified speaker (UNKNOWN): exactly right.
- 11:26
- Unidentified speaker (speaker_1): Yeah. Yeah. I
- 11:27
- Unidentified speaker (UNKNOWN): just interrupted
- 11:28
- Unidentified speaker (speaker_0): you right there. Yeah. No,
- 11:29
- Unidentified speaker (UNKNOWN): I
- 11:29
- Unidentified speaker (speaker_0): mean no, it's uh that's I feel like all of you and I's conversations over the years have involved a certain amount of like interrupt, but to build on, and then you interrupt to build on something else. So it's
- 11:38
- Unidentified speaker (speaker_1): just great riffing, right? Like
- 11:40
- Unidentified speaker (UNKNOWN): I totally agree.
- 11:41
- Unidentified speaker (speaker_1): Yeah.
- 11:41
- Unidentified speaker (speaker_0): I totally agree. No, I I think that's right. And I guess the the flip side to it, and just to circle back to the scaling point, because I am curious to your get your take on this, is I do feel like there is a sort of like trade-off that you have to make a little bit uh when you're scaling a company, growing a company, which is and I benefited from this
- 12:01
- Unidentified speaker (speaker_0): in the non non-AI AI era, the pre-AI era, which was um we're gonna hire Alex as an entry-level employee, uh, with the expectation that he doesn't know his elbow from a hole in the ground. And um, we will uh train him and be patient with him and give him projects that
- 12:21
- Unidentified speaker (speaker_0): have like a negative ROI just so he can learn. And eventually the hope is we'll come out in the black and we will be able to realize, you know, financial benefits from working with Alex. And if he stays here long enough, he'll, you know, return that investment several times over. And if not, then at least he will go off into the industry that we all compete in as a more sort of
- 12:41
- Unidentified speaker (speaker_0): knowledgeable and competent person.
- 12:42
- Unidentified speaker (UNKNOWN): Yes.
- 12:43
- Unidentified speaker (speaker_0): Uh, I benefited from that. I had an internship when I was in high school. I wasted so many other people's time and productivity learning about. this industry. That trade-off is now different, I think, in the age of AI, right? Because now if I need some entry level help and I need it to sort of start at a competency level that's much higher than Alex
- 13:02
- Unidentified speaker (speaker_0): Johnson's 16 year old intern, which is basically any unit of intelligence out there, uh, then you know, I'm maybe going gonna to do that instead. And so I'm very curious to get your perspective just as CEO of a scaling company. How do you think about the trade-off between sort of like staffing up and sort of the costs that go associated with that
- 13:22
- Unidentified speaker (speaker_0): versus AI and how you blend the two of those together?
- 13:25
- Unidentified speaker (speaker_1): Increase has been very like uh um conservative with headcount uh from the start. Part of our reason for able to do that was our structurally where we sit within the ecosystem. The tongue tongue-in-cheek in cheek here is that. it's easier to build on top of the Federal Reserve than it is to build on top of a bank building on top of the Federal Reserve. Sure.
- 13:45
- Unidentified speaker (speaker_1): And so we we kind of,
- 13:46
- Unidentified speaker (UNKNOWN): I
- 13:46
- Unidentified speaker (speaker_1): don't know, accidentally ended up with this kind of gift to ourselves where we weren't scaling with respect to to count of people anyway.
- 13:57
- Unidentified speaker (speaker_1): Like I don't uh feel good passing off to a robot yet, though. And most of them come down to like worry, like worry budget. Right. Like I I want someone to be running the like sense check level on everything. Like I I don't care that you did the the bookkeeping. I care that like you did the sense check,
- 14:17
- Unidentified speaker (speaker_1): whether or not it all looked like reasonable and sensible. I don't care whether or not you like like wrote the notes from the sales meeting. I care whether or not like like they are sensibly tied together. And we're not at outsourcing those yet. We certainly are at the like outsourced, like I've got this uh ticket sitting in my backlog. Can you run through it,
- 14:37
- Unidentified speaker (speaker_1): produce good code, or uh can you um uh produce this good piece of collateral for a sales meeting, right? Like all of those are are taken along. But the worry To outsourcing, I don't know. I'm I'm fascinated by how that will happen. I'm not sure if you have you set up a like a a series of agent advisors for yourself yet? Uh uh.
- 14:54
- Unidentified speaker (UNKNOWN): I have.
- 14:54
- Unidentified speaker (speaker_0): I have. Yeah. And it's um it's interesting because I I I totally hear what you're saying. And it it clicks sort of intuitively into my own brain and my own experiences because
- 15:06
- Unidentified speaker (speaker_0): there is just this, and it kind of goes back to what the the early stories with John and Patrick. Like there is this sort of inherent lack of urgency in AI where. it's just like it doesn't have to move fast. It will move as fast as you push it to. But like I don't feel like the best people I've ever worked with, and I'll use like um events
- 15:25
- Unidentified speaker (speaker_0): marketing people as an example. Like a great events person is the best because they are a combination of hyper competent and hyper paranoid, blended perfectly together. Right. So it's like there's they're ultra competent. They've seen everything. They know everything. They have all these little tips and tricks. They knew all these things, but they blend that
- 15:45
- Unidentified speaker (speaker_0): together with just like, you can almost feel this like tension barely under control, like vibrating off them at all times, where it's like, I never walk away from an interaction with a great events person feeling like I thought of something that they haven't thought of, worried over, lost sleep over. And in some senses, I guess
- 16:05
- Unidentified speaker (speaker_0): that's kind of like a bummer. Like, oh, humans still have to like worry about things. Like, shouldn't we be trying to remove that? But at the same time, like, I think that's what makes you really
- 16:15
- Unidentified speaker (speaker_0): not only good at your job, but valuable in your job, right? Is that like to your point, I can outsource I, one employee, can outsource my worry over this thing that I don't really have the time or expertise to cape capably worry over to someone else who does and will. And I can trust that. And like, I don't know, we we talk a lot in AI and agentech
- 16:35
- Unidentified speaker (speaker_0): stuff about like accountability and can you actually outsource this thing? Can you trust this thing? And I think you're putting your finger on a really important like subcomponent of that, which is like, can I fully outsource what I would normally worry over to this thing that can worry more capably? and more well than I can? Does that make sense?
- 16:53
- Unidentified speaker (speaker_1): It does. I'm I'm left in this like uh uh are we purposely going to introduce like anxiety to robots and like uh
- 17:01
- Unidentified speaker (speaker_1): like like
- 17:03
- Unidentified speaker (UNKNOWN): you
- 17:03
- Unidentified speaker (speaker_1): know, can we give a different feeling first, please? Uh right,
- 17:06
- Unidentified speaker (UNKNOWN): right, right.
- 17:07
- Unidentified speaker (UNKNOWN): No, it's a joy and happiness before uh anxiety. I mean,
- 17:10
- Unidentified speaker (speaker_0): you know, it is a good point because it's like you think about like the the hugging face incident. which is kind of like the magna carta of like
- 17:17
- Unidentified speaker (UNKNOWN): all of these concerns.
- 17:18
- Unidentified speaker (speaker_0): And it's like those agent swarms were clearly like desperate to please. You know, so like in some senses, I guess with these like reward architectures we're building into model deployment, we're introducing some form of desperation. But yeah, I I think we haven't quite hit on to your point, like what makes humans unique in that way,
- 17:38
- Unidentified speaker (speaker_0): relative. to AI. And so there does still feel like a gap between those two.
- 17:43
- Unidentified speaker (speaker_1): You uh have you ever had the situation with uh a work colleague, especially someone you're working with closely? Because you'd worked with them usually on in like enough projects or enough things so far, where um you just like trust that when they have it, they have it, right? Like
- 17:57
- Unidentified speaker (UNKNOWN): you're you
- 17:58
- Unidentified speaker (speaker_1): you're just like cool, you got it. And there's a really like exciting version of that, which is I'm not sure, have you ever like walked into a meeting, seen the other people who are in the meeting, and then you're like, Cool, like I like I'm adding nothing. I'm pe I'm like I'm leaving. Uh yeah,
- 18:12
- Unidentified speaker (UNKNOWN): yeah,
- 18:12
- Unidentified speaker (speaker_0): yeah,
- 18:12
- Unidentified speaker (UNKNOWN): yeah. Um like preemptively I can just opt myself out of this. Correct. Like
- 18:16
- Unidentified speaker (speaker_1): uh you you have all of my thoughts here. Like I'm not like uh uh my physical presence is not necessary for any
- 18:22
- Unidentified speaker (UNKNOWN): totally um
- 18:22
- Unidentified speaker (speaker_1): and totally and it's a really great level of trust, right? Like if you're able to reach that level of trust. with colleagues, it's just like like scales so so nicely. There were a lot of instances of that in early days strive and there are a lot of instances of that with with increase. I I've never ended up with with that with robots yet, right? Like and maybe we'll get there. Um uh uh ah you got it cool
- 18:43
- Unidentified speaker (speaker_1): I'm out
- 18:43
- Unidentified speaker (speaker_0): that's exactly the the precipice that I've walked myself to with AI agents and swarms is uh a real enhancement to sort of like the productivity of lifting and shifting and moving things around and doing things. But that level of trust still eludes me personally in terms of my use of this stuff. So I think that's I think that's a really great way to put it. All right. That is the end of
- 19:03
- Unidentified speaker (speaker_0): act one. Okay. Uh small intermission if people need to uh get up, pause for a second, and get a drink of water. All right. Uh act two in our little play is um getting close to the metal. And I want to spend some time here because when I think about the work you guys do, this is the first thing that sort of comes to mind. Um, you
- 19:23
- Unidentified speaker (speaker_0): have a design principle that you wrote about recently, actually, at Increase uh for building APIs, which is no abstractions. The piece you guys wrote on this is great. Uh, we'll link to it in the show notes. But I'm wondering if you can sort of explain that principle and why it's important. and how your experiences earlier in your career sort of
- 19:43
- Unidentified speaker (speaker_0): informed your thinking on this principle.
- 19:45
- Unidentified speaker (speaker_1): I'm going to do the thing of caveat preamble of this isn't meant to be universal, right? Like it is held true for increase. And in particular, it's held true for for for our users. Uh So um increases banking by API, the ability to open accounts, send receive transfers. issue loans, balance transaction reporting,
- 20:05
- Unidentified speaker (speaker_1): issue cards. It increases the bank that I wanted while while while building Stripe.
- 20:09
- Unidentified speaker (UNKNOWN): To
- 20:10
- Unidentified speaker (speaker_1): to contrast two uh product approaches. So Stripe product approach uh was always uh create a beautiful abstraction over payments, right? Yep. You're a developer, you want a slash payments API. Um, handle all of the complexity. Um and increases is a contrast to that of like I am an engineer at, you know, at uh Stripo
- 20:30
- Unidentified speaker (speaker_1): Ramp or Gusto, and don't hide the uh the the details from me. I want full capabilities and data of the underlying network. If it's called company entry description at FedACH, I want the parameter to be called company entry description. I don't want to have to think about how your abstraction matches to the Federal Reserve's underlying network. There are
- 20:50
- Unidentified speaker (speaker_1): ways in which it like simplifies a lot of our product approach, right? Like I'm not trying to hide the details or complexity of the underlying networks. Um, I'm just trying to pass them through with with high fidelity. And you should be able to like, you know, uh uh rely on the fact that I will do that. Then you know, when you're integrating Fed now, it should be relatively unsurprising to you
- 21:10
- Unidentified speaker (speaker_1): how I'm going to implement, you know, 2022 20 go to two uh like in the JSON API. Uh you should be like to a high degree be able to like uh uh predict what all of the parameters are going to be called and it it it it will just be you know joyfully unsurprising to you right like and that applies both to you know your initial API integration but also all of the operations. Oh
- 21:30
- Unidentified speaker (speaker_1): surely you've a hook when this is returned. Yep and surely you've the mapping for when that's returned. Great. And then you have the proof of authorization request also by API. Yep and you've got the whole hold harmless letter in the case of tertiary flows. Like, yep, all of those will will work just as
- 21:45
- Unidentified speaker (speaker_1): you as an operator who is trying to compose banking within your product would want it. Yeah, it's fun.
- 21:51
- Unidentified speaker (speaker_0): Yeah, it's interesting too, because the I think one distinction you're getting at that I hadn't fully crystallized in my own head until you just said that is there is a distinction between building for developers and building for engineers. And I am curious to get your take on this because, you know, obviously, as you said, you know, Stripe's specialty
- 22:11
- Unidentified speaker (speaker_0): is really like abstracting complexity away for developers who are trying to move fast and build a thing where they don't need to know how the underlying thing works, but they need it to work so that they can then build the thing on top of the thing. It feels to me, and I'm really curious to get your sort. of um overall perspective on this shift. It feels as though
- 22:31
- Unidentified speaker (speaker_0): over time
- 22:33
- Unidentified speaker (speaker_0): in financial services, we are moving just generally closer to the metal. And we are moving more towards a world where engineers are making decisions. And you know, I mean, I the example I would uh go back to company that I know you know well who I think thinks this way is. like ramp, right? Like they are
- 22:53
- Unidentified speaker (speaker_0): religious about letting engineers drive the decision making at that company. And, you know, it's it's really kind of fascinating to me because I think a lot of, especially in fintech, a lot of companies that are having a lot of success and scaling quickly and even as they grow, continuing to maintain a pace of shipping code and shipping products, they tend to be sort of
- 23:13
- Unidentified speaker (speaker_0): engineer engineer-led led by default. And I think, you know, infrastructure that matches up with that is incredibly valuable. I also think sometimes it can be trying to think of the right way to describe it. It can be a little bit like risky when there are engineers making decisions about infrastructure they want to use,
- 23:33
- Unidentified speaker (speaker_0): where they're only looking at it from an engineering perspective and not from other valid perspectives that are all. also important. So I'm curious to hear you kind of characterize because I've I've observed this shift and I've seen both the good and the the bad come out of it. And I'm curious to get your perspective on it.
- 23:48
- Unidentified speaker (speaker_1): But certainly abstractions done well and done well probably kind of like gets you to like not leaky, kind of free us up to higher level of thinking, right? Like uh uh uh um and those both that that's kind of powerful. It's powerful within engineering, it's powerful just more generally. The difficulty is when they they they um necessarily become leaky or you know, when you you know, there's kind of progress
- 24:08
- Unidentified speaker (speaker_1): progressive disclosure and like it it turns out that you actually need to drop down to a lower level rather quickly. So I don't want to, you know, throw the baby out with the bathwater with her speech to uh uh uh abstractions or anything. Um I think I uniquely talk about uh uh kind of the the narrower use case as it applies within uh within Increase. The
- 24:28
- Unidentified speaker (speaker_1): progress of of kind of uh of empowerment within organizations for for decision making uh
- 24:35
- Unidentified speaker (UNKNOWN): is
- 24:35
- Unidentified speaker (speaker_1): fascinating. I
- 24:37
- Unidentified speaker (UNKNOWN): kind of
- 24:37
- Unidentified speaker (speaker_1): stripe rode this uh uh wave in in kind of the 2010 timeframe when like the engineer was more uh empowered to be involved in um in service provider uh choices. Like
- 24:50
- Unidentified speaker (UNKNOWN): of
- 24:50
- Unidentified speaker (speaker_1): course your engineer was going to make the choice about whether or not you were a Amazon Web Services or a Google Cloud Platform uh uh you know shop. Um but it wasn't obvious that like they would also be involved in the decision as to whether or not you were like a you know payment tech or Wells Fargo merchant services shop.
- 25:08
- Unidentified speaker (UNKNOWN): Right.
- 25:09
- Unidentified speaker (speaker_1): Um
- 25:11
- Unidentified speaker (speaker_1): surely that one naturally maps to to to to finance function. Sure.
- 25:16
- Unidentified speaker (speaker_1): And one of Stripe's uh uh innovations was make it quick and easy for folks to be able to sign up and and and get moving. That would happen to be the engineer who was who was evaluating services and evaluating APIs. And like, you know, weeks down the line, there was the paused moment of wait, watch, who are we using for payments? That's kind of when
- 25:35
- Unidentified speaker (speaker_1): procurement would catch up and want to get involved. Um And that can be fun, right? Like uh it's it's it's an interesting uh sales conversation or you know, uh a pricing negotiation conversation when the other side has already built all the integration. You probed on whether or not uh um kind of empowered engineers have kind of unalloyed good,
- 25:55
- Unidentified speaker (speaker_1): or whether or not there's there's some places where you want to be wary. And yes, of course. Uh uh I think uh primary within this, likely within kind. of uh both of our uh professional as well as hobbies here. You'd ask very reasonably about whether or not this is kind of like unalloyed good, right? Like whether engineers' empowerment as as decision makers uh has downsides.
- 26:16
- Unidentified speaker (speaker_1): And uh I think likely relevant to kind of both of our professional experiences, uh, you want to be careful in regimes like um ensuring that the decision makers maker are is aware of compliance uh uh and regulation constraints, right? Uh uh it is there for a reason. Um, unless you know and understand the reason, please don't take it out,
- 26:36
- Unidentified speaker (speaker_1): right? Like uh uh uh Chesterson's
- 26:38
- Unidentified speaker (UNKNOWN): fence,
- 26:38
- Unidentified speaker (speaker_1): yeah. Yes. The code would be simpler if you removed the requirement for the the the data elements above you know three thousand $3,000 dollar wire transfer rule. Sure. Don't uh uh uh um without talking to your friendly compliance department, don't do that. There can also be at times, I love many attitudes about kind of traditional
- 26:58
- Unidentified speaker (speaker_1): engineering and its influence require a lot. But sometimes it can be dismissive towards the cultures or constraints of other functions. In particular, I don't know. First principle thinking is often lauded. It uh uh But first principle principles thinking can at times, if naively uh uh uh applied, uh, be dismissive of experience.
- 27:15
- Unidentified speaker (speaker_1): right? Uh I can't tell you the rule, but I can tell you that if you walk into the regulator and say this thing, they are immediately going to say, That's nice, get out now.
- 27:25
- Unidentified speaker (speaker_0): Just one of those immutable, mysterious laws of the universe where it's like you'd have no reason to know this, you can't intuit it from first principles, but trust me, it's there.
- 27:33
- Unidentified speaker (speaker_1): It's a reasonable thing to expect experienced practitioners to call out which ones are enumerated and which ones are experienced. experience based and which ones, you know, where are you spending against your innovation budget. And as long as you all are able to know, know and recognize those concepts, speak them out loud, trust that the good intentions of others, trust the competence and experience and uh uh uh goodwill
- 27:53
- Unidentified speaker (speaker_1): of others, you can do it well. It that's actually easier to do with a smaller team though than a larger team. Um
- 27:57
- Unidentified speaker (speaker_0): that that I mean, that makes a ton of sense to me. And I guess on your point about intentions, the other thing that I think is interesting to kind of press on there is there is this weird,
- 28:07
- Unidentified speaker (UNKNOWN): like
- 28:07
- Unidentified speaker (speaker_0): sometimes financial services is this sort of like semi-permeable barrier where it's like, yes, that is a wall, but it's kind of hard to tell from the outside if that's a wall that you can actually phase yourself through with kind of consistent applied effort, or if it's a wall where no matter how hard you press against it, you're never getting through that wall and you're wasting your time and
- 28:27
- Unidentified speaker (speaker_0): energy. And, you know, like compliance, obviously, and risk plays a big role in that, right? Like if they go to you and they say, hey, like physically can this be done? Can money move in this way? Can we settle this this way? Can the product construct work this way? Yes. According to the physical laws of the universe, it can. Is that legally allowed? No.
- 28:47
- Unidentified speaker (speaker_0): Is it possible that you could do something kind of similar to it, but different? in these five ways? So I like I guess what I'm getting at is a lot of the value in sort of applying technology to financial services is also, I think, applying a certain amount of like stubbornness to like which of those barriers are actually impermeable and which ones are sort
- 29:07
- Unidentified speaker (speaker_0): of semi permeable if you push in the right way and with the right type of effort and with the right technology. Do you have any kind of like lessons? learned or observations about how to kind of strike that balance? Cause I do go back to the word you mentioned earlier in like good intentions, right? Like to me, a lot of times it's almost a downstream effect of
- 29:27
- Unidentified speaker (speaker_0): if we are trying to innovate for the customer and we're trying to do it responsibly, and this barrier we keep running up against just doesn't make any sense and feels like one we should be able to push past. Like it feels like there's a relationship between those two. But obviously, like a lot of what we're describing is very case by case. There's not like clear and fast rules. How do you think about navigating that tension?
- 29:47
- Unidentified speaker (speaker_1): Different organizations are going to do this differently. And one model, you know, uh is is set up antagonism. And that that can be fine in certain places, right? Like literally, we purposefully have set up antagonism within society, uh uh uh, you know, within like legal system. It's hard for that one well to work. well inside in a company though, right?
- 30:05
- Unidentified speaker (speaker_1): Like it needs to feel like a, you know, shared sense of struggle. It needs to feel like uh uh us against the problem and we're both sitting at the same time side of the table. And so the like, you know, eking out concessions from your, you know, compliance or legal function is is it's certainly like uh uh uh uh be wary. You can end up building yourself a
- 30:25
- Unidentified speaker (speaker_1): lot of of things that look like uh technical debt there. You may or may not want that, right? Like you may or may not want to have have solved it differently. Um uh yeah, and the kind of respect and uh attribution of of professionalism to different places within your organization that are not, you know, uniquely code contributors can be very
- 30:45
- Unidentified speaker (speaker_1): useful. Um I I I guess uh this is one of the the things I'm like most excited about what a uh a like the robots inside and companies help us with, right?
- 30:56
- Unidentified speaker (UNKNOWN): Because
- 30:57
- Unidentified speaker (speaker_1): no one is there as a warm body anymore, right? Like someone is there as an expert, right? Like no one is there as a like ah you're just picking
- 31:03
- Unidentified speaker (UNKNOWN): the sheet of the
- 31:03
- Unidentified speaker (speaker_1): spreadsheet, right? You truly value the expertise of of everyone who is there um yeah because they're that worry bearer darer for you. Right. Like they're they're uh yeah Um that's
- 31:13
- Unidentified speaker (UNKNOWN): a good way.
- 31:13
- Unidentified speaker (speaker_0): uh No, that's a good way of putting it. I mean, because I do think like if I was to sort of empathize with people who sometimes maybe don't listen inside organizations to other legitimate concerns, it there is this almost element of like, you're just a human in my way trying to stop me from doing this thing, not someone who's on my same team,
- 31:33
- Unidentified speaker (speaker_0): who's here for a very valid reason, who has expertise I don't have, and to a degree it does feel like our need in a pre AI era to just have bodies and seats to do jobs did contribute to a little bit of that kind of lack of shared values or like antagonism that would develop between different groups, silos,
- 31:53
- Unidentified speaker (speaker_0): however you want to describe it. Whereas I do think in the future, and we're in the midst of this change, so it'll take a little while for this to play out, but we probably will see like a different sort of status quo settle out where it's like my default assumption is if I'm interacting with a human on this, it's because this human is there for a very valid reason and I should like
- 32:12
- Unidentified speaker (speaker_0): empathize and have respect for that. I'll f I'll save my fighting for fighting with the robots, but when I'm working with a human, we're we're we're on the same side here.
- 32:20
- Unidentified speaker (speaker_1): Yes. Yes. It's uh uh uh you know, I won't interrupt a human as much as I would the robot.
- 32:25
- Unidentified speaker (speaker_0): Yeah. Exactly. That's exactly right. The other part of this that um we haven't mentioned yet, but we should talk about is uh increase is becoming a bank, uh, which is very exciting. Um I I'm I know I'm phrasing that a little imprecisely, so feel free to uh phrase it more precisely now that you're living in this world. But uh I I know
- 32:45
- Unidentified speaker (speaker_0): this has been sort of directionally um something that you've been interested in for a while. So maybe just talk. to us a little bit about the in the in the language of removing abstractions,
- 32:56
- Unidentified speaker (speaker_1): the
- 32:56
- Unidentified speaker (speaker_0): benefit of a bank chart.
- 32:58
- Unidentified speaker (speaker_1): Yeah. Yes, I had started increase, you know, six, seven years ago with the intention of being a global bank with a technology muscle. That's what I wanted while building Stripe. That's uh uh um and kind of prerequisites for that are you have to have the technology muscle, you have to have uh uh uh licensing. It took us a few. years, but um maybe about a year and a half ago,
- 33:18
- Unidentified speaker (speaker_1): uh, we purchased a bank. Um, it's since been rebranded to Increase Bank. Um, it took us a while to put the correct processes and technologies so that we could operate safely and soundly the way kind of Increase uh uh wants to operate. That's fun. For years and years and years, we were already integrated in at kind of the the lowest technology level, right? Like
- 33:37
- Unidentified speaker (speaker_1): interacting with the federal. Reserve and kind of moving the electrons and the sense around. Um, but stepping across to the charter institution side is is fun and fascinating also. You know, it comes with all of the things that you'd expect, right? You know, asset liability management, uh liquidity management, uh um communicating
- 33:57
- Unidentified speaker (speaker_1): all of what you're doing to the examiners, um, ensuring that everyone is aligned. and eyes open to everything that's happening and what the bank is able to do and not able to do and uh uh where the correct paper trails needs to be. Um but none of that was you know, we we we didn't shy away from that at any time during increase's history like
- 34:17
- Unidentified speaker (speaker_1): before the bank. Uh uh so I think it's it's it's it's just like a natural addition to what it is that we do.
- 34:23
- Unidentified speaker (speaker_0): That makes sense. Is there are there specific things that um will be easier or different having the bank? Because I mean, I I spend quite a bit of my time, uh, as you know, sort of puzzling over banking as a service and embedded finance and all the different models that exist here. And you know, it's interesting because there's a ton of variability, right? I mean,
- 34:43
- Unidentified speaker (speaker_0): um so-called middleware platforms that would sit between end users of services and regulated banks. Million different models, million different ways for that to work. Some were very focused on abstraction, to the point of extreme danger and risk and problems that would creep in. And then there were others that were trying to be very much
- 35:03
- Unidentified speaker (speaker_0): just like uh passing through at the highest possible fidelity. So there's differences there. There's also, I think, a lot of differences in terms of, you know, the banks in question who hold the charters, how well they work with FinTech. company. So there's a million different variations. But what are some areas where you think this might create additional kind of leverage or opportunity
- 35:23
- Unidentified speaker (speaker_0): for you guys?
- 35:23
- Unidentified speaker (speaker_1): So some comes down to the product experience that you can create for for for users. You just like uh uh you're at times can be more culturally aligned. Um if you'll forgive a story, Stripe was built on top of Wells Fargo. At one stage there's a a piece of uh data that traverses the FedACH network called the trace number, uh a uh
- 35:44
- Unidentified speaker (speaker_1): not quite unique identifier for each transfer. And because of the technical integration with Wells Fargo, the the identifier as generated by Stripe would not be the identifier that was like returned back during a return. And therefore the correlation of the originations to the returns uh uh uh was challenging. And uh communicating this to Wells Fargo. was like it was
- 36:04
- Unidentified speaker (speaker_1): was hard. And no fault
- 36:05
- Unidentified speaker (UNKNOWN): on
- 36:05
- Unidentified speaker (speaker_1): Wells Fargo, it was just built for something else, right?
- 36:07
- Unidentified speaker (UNKNOWN): But
- 36:07
- Unidentified speaker (speaker_1): like uh uh um you kind of felt like
- 36:11
- Unidentified speaker (speaker_1): like Princess in the P, right? Like
- 36:14
- Unidentified speaker (UNKNOWN): you're you're like
- 36:14
- Unidentified speaker (speaker_1): you're you're pointing at this thing and everyone's like, ah, surely that's a small problem. And you're like, no, I've 12 people doing this thing because of like lack of this piece of information. Totally. You're just wearing someone else's shoes. right? Like
- 36:27
- Unidentified speaker (UNKNOWN): you're just
- 36:27
- Unidentified speaker (speaker_1): you're you're you're you're uh not with your people or something's, you know,
- 36:31
- Unidentified speaker (UNKNOWN): something's a little uncomfortable.
- 36:33
- Unidentified speaker (speaker_1): And getting to be the bank that's built for ambitious technology muscles uh uh or ambitious technology companies is great, right? Like you you oftentimes in a like a user conversation uh come to a moment where they ask the like the question that reveals some sort of scar tissue. And when you like echo it back and you're like, yep, seen
- 36:53
- Unidentified speaker (speaker_1): that. Here's how we handle it. And suddenly there's this like,
- 36:55
- Unidentified speaker (UNKNOWN): I
- 36:56
- Unidentified speaker (speaker_1): know, sigh, relaxation of the shoulders, some
- 36:59
- Unidentified speaker (UNKNOWN): sort of like,
- 37:00
- Unidentified speaker (speaker_1): I'm seen, right? Like, great.
- 37:02
- Unidentified speaker (UNKNOWN): I'm
- 37:02
- Unidentified speaker (speaker_1): I'm not going to have to like check under absolutely everything here. Uh uh at all. Because you've lived some of this life with me and and and uh yeah you're not munching the trace ID. Great, cool. Um
- 37:15
- Unidentified speaker (speaker_1): we
- 37:15
- Unidentified speaker (UNKNOWN): can
- 37:15
- Unidentified speaker (speaker_0): spend our our precious resources on other things. Yeah, yeah.
- 37:19
- Unidentified speaker (speaker_1): Uh uh yeah, I'm not gonna have to, you know, sniff around the room for the thing that's that's off. So some of that is fun, right? Like some of that is very much just like uh uh banks were always meant to be like aligned with user needs, right? Like
- 37:32
- Unidentified speaker (UNKNOWN): we
- 37:32
- Unidentified speaker (speaker_1): just didn't have ones that were aligned towards technology companies. We had
- 37:35
- Unidentified speaker (UNKNOWN): ones that
- 37:35
- Unidentified speaker (speaker_1): were aligned towards uh banking in Somerville, Massachusetts, but we didn't have ones. that were aligned towards banking in like four ambitious technology companies. So that's fun. There's other parts which are product velocity helping. And some of this is uh I don't know, you'll laugh at this one. Fed now request for payments, right? Like good luck making a competent
- 37:55
- Unidentified speaker (speaker_1): business case as to how this you you should spend technology like budget on this. Absent being able to point towards marketing budget, you're not going to be able to. Like there's no spreadsheet that this comes out cost benefit positive for you. And so, you know, if if if I'm showing up at someone else's bank saying, Yeah, I I really need this because
- 38:14
- Unidentified speaker (speaker_1): technology forward companies look for things like this as signaling that you like, keep up, right? That you're like, you know, I I'm I'm I'm not gonna have to hound you on your your your roadmap. But when it's our own bank, like we don't have to
- 38:27
- Unidentified speaker (speaker_1): I I don't have to show you the spreadsheet for we all know that we're doing this for marketing benefit for now. In a year's time, we'll be happy that we did it. But uh uh uh so some of those are real near-term user experience. Some are kind of literally better safety and soundness. Uh um, you're the kind of company that has this kind of liquidity profile. Yep, we know
- 38:47
- Unidentified speaker (speaker_1): how to handle that. We're not going to like accidentally equate the that with the law firm or the local doctor's office in their liquidity profile. So some is genuinely just like better, like nuts and bolts banking. And then some of it is more kind of like speculative marketing efforts that can nod towards who you want to be in the future in a different way.
- 39:04
- Unidentified speaker (speaker_0): This episode is brought to you by Increase. Increase offers API-first API First banking for ambitious technology companies. Companies like Coast, Gusto, and Ramp rely on Increase Technology to process more than $500 billion annualized in payments, card transactions, and loans. The founder of
- 39:24
- Unidentified speaker (speaker_0): Increase was Stripe's first employee, and he's built a team of industry veterans who understand technology and banking better than anyone. If you're building the next great financial product and need a bank that's designed for your needs, see for yourself at Increase.com. increase.com.
- 39:41
- Unidentified speaker (speaker_0): That's a good
- 39:43
- Unidentified speaker (speaker_1): and
- 39:43
- Unidentified speaker (speaker_0): comprehensive list. Cause yeah, it's it's always interesting. I mean, there's I sometimes think about banks as sort of like a different aggregation layer for in the stacks, right? And so it's like if you're a bank and you're working with various fintech companies or middleware platforms or what have you, like in aggregate, everything they want needs to be the thing that drives your. priorities.
- 40:03
- Unidentified speaker (speaker_0): But all of them might be slightly different from each other. Or in the same way, like I always thought of it as kind of like uh a negative selection business in some ways where it's like, you know, we have to care a great deal about who all their other customers are because if there's one bad Apple in there, it kind of spoils the whole bunch. And so there's
- 40:20
- Unidentified speaker (UNKNOWN): oh
- 40:20
- Unidentified speaker (speaker_1): agreed.
- 40:21
- Unidentified speaker (speaker_0): Getting getting a layer lower uh just simplifies a lot of that. Um, all right, now in the remaining time we have act three, we're gonna go fast. We're gonna give ourselves a time limit here. I say we do 10 minutes, as many AI questions as I can get you to go through. Are you ready, Dara?
- 40:38
- Unidentified speaker (speaker_1): I I I think we can do this together. Yeah.
- 40:40
- Unidentified speaker (speaker_0): I
- 40:40
- Unidentified speaker (UNKNOWN): think so too.
- 40:41
- Unidentified speaker (speaker_0): Okay. We're gonna we're gonna get to consensus quickly. All right, question one. How bullish are you on agentic commerce? And part two, are there specific agentic commerce use cases that you are more or less high on than consensus?
- 40:53
- Unidentified speaker (speaker_1): Short answer up front, bullish. Uh uh, but the things I guess I'm most uh excited about are I'm not sure have you granted yourself the gift of giving your agent a payment card yet? If you have not,
- 41:05
- Unidentified speaker (UNKNOWN): uh
- 41:05
- Unidentified speaker (speaker_1): wish it upon you. Uh
- 41:06
- Unidentified speaker (UNKNOWN): I have. I
- 41:07
- Unidentified speaker (speaker_0): have.
- 41:08
- Unidentified speaker (speaker_1): Uh have you given it your daily driver payment card, or do you give it an isolated device? I have
- 41:14
- Unidentified speaker (speaker_0): I've given it the isolated uh one for now.
- 41:17
- Unidentified speaker (UNKNOWN): I
- 41:17
- Unidentified speaker (speaker_1): think that's the correct way of going about it. I think it's fun to be able to uh uh provision a like payment instrument for uh, you know, that can be a whole bank account if you want, it can be the account numbers, it can be uh uh payment card itself. Um and kind of granting it the uh the economic agency that should go with the the the the rest of being an agent.
- 41:37
- Unidentified speaker (speaker_0): Next one. Um and then related. So maybe you can uh expand on your answer a little bit more. Do you think AI and Agentic Commerce will help or hurt small businesses and small merchants?
- 41:47
- Unidentified speaker (speaker_1): I think this is one of the ones I'm most enthusiastic about. I, like presumably uh uh most people, have aspirations for more local commerce, right? Like to uh I'd
- 41:57
- Unidentified speaker (UNKNOWN): like to buy it
- 41:57
- Unidentified speaker (speaker_1): from the local business, right? Like I'd like to buy from the local coffee roll store, street. I'd like to buy from the local bookstore. At times we end up time deficient or stressed and we just like, cool, I'll guess I'll get that off Amazon right now. Yeah. Um so you know, our revealed preference is certainly different from from if you if you if you ask me what I was hoping for. Um but in a world of more
- 42:17
- Unidentified speaker (speaker_1): ability to offload that, um, asking the agent to go order from, you know, order uh the the beans from thump coffee up the road. is great. Asking it to order from roundabout books down the road is fantastic. Its ability to drive like we could be at the point where this gets to drive more commerce that's more aligned with
- 42:37
- Unidentified speaker (speaker_1): our um who we aspire to be than who we practice. And that'd be fun, right? Like uh
- 42:43
- Unidentified speaker (UNKNOWN): if
- 42:43
- Unidentified speaker (speaker_1): we're able to use that to promote local businesses, if we're able to use that to promote for businesses in small towns. Like I don't know that there's a very hopeful version of that.
- 42:50
- Unidentified speaker (UNKNOWN): I
- 42:50
- Unidentified speaker (speaker_0): like that I will editorialize on your answer for 30 seconds to say that uh everything you just described pretty much describes the difference in relationship uh with Muse between Shopify and Amazon. Indeed. And so uh we're seeing that that premise play out right now in real time. And uh I will also say that there is something very
- 43:10
- Unidentified speaker (speaker_0): cool about the like long tail of commerce searches that I've never really felt fully satisfied on, where
- 43:16
- Unidentified speaker (UNKNOWN): it's
- 43:17
- Unidentified speaker (speaker_0): like, I will go in and I'll type out like this is what I want. And like I describe it in this like exacting amount of detail. And then I get the most generic possible sorting of that. Oh, okay, you want an axe. And I'm like, no, that's not what I said. What I said was I want a 36 long, 36 inch
- 43:37
- Unidentified speaker (speaker_0): long felling axe, and it needs to be forged in one of these three uh bled blacksmiths in Sweden and it needs to have this handle like I'm I'm describing something very specific. And it turns out, Dara, that there is one forge in Sweden that meets all my requirements, but they only make like 150 of the damn things a
- 43:56
- Unidentified speaker (speaker_0): year. And I want one, and I'd like to get in line and I'd like to find that. And it's like I I I feel like I'm I almost all of the things I care about are these really obscure, long tail, hyper specific, well-crafted things that I want. And now I'm very excited that I can just ask, I can describe my feelings to AI and it can
- 44:16
- Unidentified speaker (speaker_0): translate that into a search that will return something like that. Early days, but I'm excited about that.
- 44:21
- Unidentified speaker (speaker_1): How much do you enjoy that shopping process for something like that? Right? Because
- 44:24
- Unidentified speaker (UNKNOWN): something like that
- 44:24
- Unidentified speaker (speaker_1): seems like it's gonna be, you know. uh uh uh you'll enjoy telling stories about that commerce process, right? Like
- 44:32
- Unidentified speaker (UNKNOWN): that's
- 44:32
- Unidentified speaker (speaker_0): exactly right. That's
- 44:33
- Unidentified speaker (UNKNOWN): exactly
- 44:33
- Unidentified speaker (speaker_0): right. And I I will um add to your point about uh revealed preference versus stated preference. My uh stated preference and what I hope eventually becomes my revealed preference is I'm happy to spend more on that experience, right? So
- 44:44
- Unidentified speaker (UNKNOWN): like
- 44:45
- Unidentified speaker (speaker_1): Yeah.
- 44:45
- Unidentified speaker (speaker_0): I feel like this stuff can really like thrive economically, even in the age of AI, if we can match, if we can do better matching, basically. Like help me find the thing that I am willing to pay way above market rates for because I value it differently.
- 44:59
- Unidentified speaker (speaker_1): Yeah. Yeah. That'll be fun. I'm laughing most because uh uh uh your example of you know an artisan axe is vastly different than my example of like my five-year-old five year old lost her headband at the soccer game. And uh I I I I need to find a replacement for this now. please. Uh uh
- 45:17
- Unidentified speaker (speaker_1): Talk
- 45:17
- Unidentified speaker (speaker_0): about, I mean, talk about a great use case where it's like, you need to save me from my five year old having a meltdown by like like as quickly as possible. There's there is no greater motivating force in the universe. Um, all right, let me ask you a couple more. Um this one, very relevant to you guys. What's one thing that AI agents are going to need from
- 45:37
- Unidentified speaker (speaker_0): banks that they may not be? getting right now?
- 45:40
- Unidentified speaker (speaker_1): I think some of this is like isolated accounts, right? Like you you just said to me that like you're giving it not your daily driver card, but your um uh uh uh something more isolated. I prefer that as my banking mechanism of choice versus say uh a bunch more control planes or a bunch more approval planes. Sure. Sure. Uh
- 46:00
- Unidentified speaker (speaker_1): um approval planes detract from my attention, uh uh, but just giving it like here's one hundred dollars an account, uh uh uh boy do I hope you get this right rather than ordering me fifty apples. If I asked, you know, the average person on the street to oh go go find me a bank that's going to let you provision different payment cards, uh uh uh uh instruments, unless you're one
- 46:20
- Unidentified speaker (speaker_1): of a half dozen banks in the nation, you you're you're you're not, right? Uh
- 46:24
- Unidentified speaker (UNKNOWN): Yeah.
- 46:24
- Unidentified speaker (speaker_1): uh uh so banking certainly needs to to kind of help consumers in that way. I worry that we end up thinking about it somewhat incorrectly though. So uh uh uh uh oftentimes within banking conversation it's it's um how can we increase like uh uh uh trust sufficiently um uh and kind of my I think falsifiable prediction
- 46:44
- Unidentified speaker (speaker_1): here is that it will have very little with actually increasing trust is what drives like adoption of these agencies uh uh agents. agents, as opposed to uh um them just sufficiently becoming convenient. I think our trust level will in fact remain the same. Our just like willingness to trust at that level will uh cross over with convenience. Um and therefore, if what
- 47:04
- Unidentified speaker (speaker_1): we're trying to do is instill more trust, certainly sure, go for it. But for me, it'll actually be driven by uh by convenience.
- 47:11
- Unidentified speaker (speaker_0): Yeah, I mean, the relationship between this thing just works. and works well and trust, I think, is something that we don't talk enough about. Like sometimes we treat these as like completely separate attributes rather than like, I mean, honestly, like what is trust if not just value that's delivered consistently to me uh over time, you know? And like
- 47:31
- Unidentified speaker (speaker_0): that's what that's what trust emerges from. And I I do I will say as a closing note that most banks still seem to be maybe thinking about AI agents more in the Amazon context than in the Shopify context. Uh, and I do think that that is an opportunity for the banks or
- 47:51
- Unidentified speaker (speaker_0): non-banks that work with a bank that thinks this way to sort of define and build different experiences than what we have today.
- 47:58
- Unidentified speaker (speaker_1): I feel like we're gonna going to end up referring to this model quite often of the Shopify model versus the Amazon model. Uh So thanks for introducing that within our uh within our Lexicon.
- 48:06
- Unidentified speaker (speaker_0): It's my pleasure. And that uh takes us to the end of our 10 minutes. I think we did very well there, Dara, especially given that you and I really like to talk. Um and uh as always, uh it was a pleasure to talk to you. Thanks for coming on the FinTech Takes Podcast.
- 48:19
- Unidentified speaker (speaker_1): Mate, Cheers, thank you for having me.
- 48:21
- Unidentified speaker (UNKNOWN): Of
- 48:21
- Unidentified speaker (speaker_2): course.
More from Fintech Takes
Fintech Takes x Nova Credit Presents: The Car Lot Meets Cash Flow
Episode discussion: DealerTrack integrates Nova Credit cash flow data to automate income verification and reduce fraud in auto lending
Speakers’ claims: Andy Mayers explains that DealerTrack is integrating consumer-permissioned cash flow data via Nova Credit to replace easily faked pay stubs with verified income data. This integration aims to automate the 16-18% of applications requiring manual income validation, speeding up the lending process. Mayers notes that dealers only see verification completion, not sensitive data, while lenders gain accurate debt-to-income ratios for better pricing. Speakers discuss potential future integrations earlier in the consumer journey and the passive adoption model for dealers.
The Blue Wave Hits Financial Services
Episode discussion: Midterm blue wave shifts financial oversight to Democrats targeting crypto and trust charters
Speakers’ claims: Host Alex Johnson and guest Rob Blackwell analyze the 2026 midterm elections, predicting Democratic control of the Senate and House will lead to aggressive financial regulation. Blackwell argues the election was driven by voter concerns over inflation and corruption, alienating the crypto industry which spent heavily against Democrats. The speakers expect Senators Elizabeth Warren and Maxine Waters to lead hearings on entities like World Liberty Financial and scrutinize OCC and FDIC actions. They predict low bipartisanship odds due to political conflicts and note state-level populist agendas targeting banking fees.