greetings islanders,
this edition of island fintech is a bit of a unique one. stripe, a company we admire here at island fintech, has picked the two hardest problems in payments, and their solutions share an interesting kinship. last week, the team were hosting stripe tour in singapore. we were invited to dinner with sophie and felix, who lead a lot of stripe’s thinking on stablecoins and agentic payments.
it’s informed a lot of our special dive this week, and we hope you enjoy it!
if this is your first time reading, island fintech has been covering fintech in southeast asia for nearly 5.5 years now (launched may 2021).
we aim to provide a refreshing lens in this region. subscribe and stay connected:
🤿 special dive: intentional agents and stable airspaces
asking for intent
just imagine, what could go wrong? you’ve asked an agent to organise a “memorable date night”. using your card, they’ve found a way to get a booking for a lavish, prix fixe meal... certainly memorable, but not what you wanted.
the problem is, no bank will catch this, because the card was authorised and the merchant is legitimate. this is your card, and it was used exactly as instructed. the only problem is, the intent wasn’t well understood by the agent.
banks don’t screen for intent, though.
could entitlements solve this? anyone with an expense card will know how this works - a person gets a defined scope for what they can spend and approve. but consumer finance has no equivalent, especially if you want some level of flexibility for your spend.
and even with limits in place - how can you prove your agent made the purchase? can you say it got the order wrong and be believed? can the merchant trust you won’t say it every time? there will always be an argument to be made on either side.
the solution for incorrect intent is therefore a ctrl-z switch - and reversal isn’t some exotic technology. it’s a couple of things - 1) mandates - a signed record of what you asked for 2) a credential tagging it as agent-made, and 3) special agentic dispute rules. visa, mc and amex have all been moving, with amex going furthest - guaranteeing that they’ll give your money back. if the mandate is clear, then that michelin dinner is your problem. outside it, you get your money back.
entitlements are essentially guardrails, and reversal saves you when they don’t work. agents need both. however, the reversal is harder to pull off, because someone has to take the hit on the refund.
when it comes to agentic commerce, there’s a big catch. all of these dispute rules live inside unique, singularly defined rule books (visa's intelligent commerce connect, launched in april ‘26, mc’s agent pay acceptance framework, oct ‘25). if you’re trying to cohere this with domestic payment schemes, don’t expect any of the rules to hold anymore. welcome to part II.
in your atmosphere
money operates across two separate airspaces, if you will.
at altitude, fiat and payment rules at country and central bank level are pretty clear. the participants are licensed, boundaries are drawn. singapore has our fast standard, the uk has fps... and so on.
when you descend to a different altitude - that space occupied with merchants, networks, issuers and acquirers operate, things look different. you’ll find totally different rulebooks, different scheme rules, different alignments. there’s no shared map.
I visited shanghai recently, and google maps doesn’t work in china - which I learnt was not just due to the google ban, but because china runs on its own gcj-02 geospatial standard instead of the global wgs-84 system. same planet, different map. that’s roughly the situation in payments.
so the thing is - both airspaces work on their own terms. very little governs the passage between them! it would be rather helpful to agree on rules - specific enough to enforce, and yet, consistent enough to rely on.
but agreeing on rules is hard. and that’s why a consortium, a team of actors with shared goals, beats a single champion.
this is where libra once came in, and where open usd now does. libra was directionally correct, but very different in execution. it started outside the perimeter. a proposed digital country, pushed forward by facebook, with <20 half-hearted members who left before the first libra meeting in oct ‘19. perhaps they were too early, or perhaps the industry had not yet caught on.
open usd, announced in june ‘26 by open standard with more than 140 partners behind it and slated to go live later this year. this starts inside the perimeter: visa, mastercard, amex, blackrock.. even dbs and ocbc were among them.
the idea didn’t improve, but the airspace got clearer.
bridge, acquired by stripe in early ‘25, follows similar thinking - with payout capability through stripe’s global disbursement stack, country coverage is stacked. and the thread isn’t accidental! open standard is led by zach abrams, the co-founder of bridge. the person who built the low-altitude licensing stack is now running the cross-altitude consortium.
it also makes the payment airspace much more navigable, and not purely on the merits of settlement assets. it’s the licensing, the local coverage and the market-by-market rules already absorbed, which is exactly the low-altitude work most stablecoin projects have avoided. stripe did the boring part first, same as it did with cards.
stripe’s openrouter deal, reportedly over $8bn, is a different bet in the same direction: one key and one endpoint in front of many models, the same shape stripe already built for money. every builder will soon need two utilities, a money pipe and a model pipe. stripe owned one and has bought a position in the other.
what do agents and coins have in common?
agents need scope, accountability and a way back. stablecoins need rules that hold across very different sets of participants.
both are areas that concern who is answerable, and get solved by mutually-aligned governance, licensing and consumer trust.
it’s indeed easy to overlook the small and boring pieces holding everything together. but that is truly where the opportunity sits. when you snap a limit on what your agent may spend. having a clear record you can read. transaction rules that still apply when the money changes altitude, whether it’s across networks or currencies.
the resultant trust improvements have the power to change behaviour more than you’d expect.
my bet is that agentic rulebooks will likely fracture at the domestic scheme level. visa and mastercard have each written their own rules - being the master of card schemes. but they don’t have any direct access into clearing on domestic schemes. the first serious agent dispute that crosses a domestic scheme boundary will cause things to break, and it’s likely a regulator writes those rules before either network does.
at today’s pace of expansion, everyone seems focused on moving as fast as they can. however, at speed, the rivets matter even more than the engines.
Island Fintech attended a dinner hosted by Stripe in Singapore in August. No conditions were placed on this piece.
dips 🎣
1. visa and nium settle live on singapore’s stablecoin rails
what happened: on 25 august, visa joined BLOOM (borderless, liquid, open, online, multi-currency), the MAS-led settlement initiative, with cross-border payments firm nium as its first pilot partner. the pilot tests whether regulated USD- and EUR-backed stablecoins can settle cross-border payments seven days a week, including weekends and public holidays, when conventional settlement windows are shut. visa is the first major card network to join BLOOM, which MAS launched in 2025 on the back of project orchid. circle, DBS, OCBC, partior, stripe, UOB and, recently, maybank singapore are also in.
🏝️ if insight: this is the direct sequel to the citi tokenised-deposit story from our july edition. citi proved a single bank could move dollars over the fourth of july; visa and nium are testing the same outcome across an open network rather than one bank’s balance sheet. the pattern is now hard to miss - weekend and holiday settlement gaps are being treated as a defect to engineer out, not a fact of life. if your settlement engine still stops on saturday, that is a competitive gap, not an industry norm.
🔗 reference: TechNode
2. the real cost of singapore’s digital bank race
what happened: the 2026 singapore digital banking report laid out FY2025 results for the city’s five digital banks.
GXS posted the heaviest loss at S$132 million, though it narrowed for a second straight year (from S$145 million)
maribank lost S$55.6 million even as income rose 53%, and grew deposits to S$1.9 billion, second only to trust
trust bank lost S$53.5 million, narrowed its loss 42%, and hit its first profitable month in march 2026
anext widened its loss 34% to S$49.8 million, and was the only one whose loan book shrank
green link digital bank (GLDB) was the lone profit at S$16.1 million - the first singapore digital bank to turn a full-year profit, built entirely on SME supply-chain financing with no retail customers
🏝️ if insight: four years in, the field has split cleanly. you either sit inside a consumer ecosystem big enough to absorb acquisition costs - grab and singtel behind GXS, sea behind maribank - or you pick a narrow, high-margin niche and actually make money, as GLDB did in supply-chain finance. the one to watch is anext. in june it moved into GPU financing, lending against AI hardware as collateral. a digital bank treating compute like bankable real estate tells you where it thinks the next wave of credit demand sits.
🔗 reference: Fintech News Singapore
3. funding hits a decade low, and MAS steps in with S$220m
what happened: KPMG’s pulse of fintech H1’2026 put singapore fintech investment at just over US$499 million across 53 deals - down from US$1.45 billion across 97 deals a year earlier, and the weakest first half in close to a decade. a single US$320 million cross-border payments round in june made up nearly two-thirds of the total. three days later, on 31 august, MAS unveiled FSTI 4.0: a S$220 million commitment over three years, 47% up on the last round, across six tracks weighted toward frontier tech and talent. it will co-fund internship stipends for at least 1,000 students through a new SFA-run portal, fintechinternships.sg, and a new AI pathfinder track will subsidise financial institutions adopting market-ready AI tools.
🏝️ if insight: read the timing. MAS is looking at a venture market that has retreated to early-stage bets and stepped in as the ecosystem’s bridge financier - but the money points squarely at AI adoption and talent, not growth for its own sake. two practical (budget friendly) reads for our SG based early-stage founder readers. the cheapest headcount you can add right now runs through the subsidised intern pipeline. and if you sell AI tooling to banks, being an approved solution under the AI pathfinder track turns a MAS grant into your best enterprise sales pitch: your buyer’s budget is partly underwritten by the regulator!!
🔗 reference: MAS | KPMG via TechNode
some further dives 🤿
four resources worth your time this edition:
FSTI 4.0 scheme details - MAS. the primary source on all six funding tracks, from centres of excellence to infrastructure builds. your grant playbook for the next 36 months. Read here
fintechinternships.sg - SFA. the portal behind the 1,000-intern pipeline. if you are hiring junior talent, register early before the neobanks take the best of the cohort. Visit portal
pulse of fintech H1 2026 - KPMG. the maths behind the squeeze. worth noting: digital assets and crypto led on deal count even as total dollars fell, while one payments round carried most of the value. Read here
2026 singapore digital banking report - fintech news singapore. the full FY2025 breakdown. one figure to sit with: maribank’s income now covers close to two-thirds of its annual loss, against 4% a few years ago. Read here
if you’re building on any of the above - 24/7 stablecoin rails, compute financing, or even the new FSTI 4.0 grants - reply and tell me what you’re seeing. the best material here usually comes from you. reach out!
until next time, island fintech