SIGNALSPowered byALFIN
SFF 2026 · Day 1 · 18 November 2026

The Daily Rewire

Day 1 in review: the shifts that mattered across 85 sessions, in five minutes.

85 sessions reviewed

These takeaways are the SIGNALS desk's editorial reading of the day's discussions. They are not commitments agreed by participants. Figures are as said on stage.

Pick a Rewiring

Sessions set near-total dollar denomination of stablecoins against euro ambitions, and argued that export controls stop chips but not open AI models. Regional blocs and payment schemes moved to link national rails while keeping sovereignty, with bilateral links straining at scale.

Macro, Geoeconomics & Talent

  1. Stablecoins remain almost entirely dollar denominated, even in Europe where the rules favour euro coins.

    Speakers cited 99% of the European stablecoin market as dollar-denominated, mostly because crypto trading liquidity sits there, and another session put the dollar share at 99.5%. The euro carries 20% of global payments but less than a third of one per cent of stablecoins, while the dollar holds more than half of official foreign reserves.

    Speakers split on the remedy. One argued for a pro-stablecoin policy to support the euro's international role, pointing to demand in South America, Africa and Asia to diversify from the dollar. Another expected dollar coins to dominate for four to five years while monetary controls and policy gaps are resolved.

  2. Export controls can stop chips at the border but not downloadable models, shortening their useful life.

    One address said that within three years of the first US controls on advanced chips, Chinese labs released open models close to the frontier. Open-weight models from China were described as spreading fast in emerging economies.

    The counter-case is concentration: the stage heard of an AI supply chain with almost no redundancy, from one maker of the most advanced lithography machines to three hyperscale clouds. One proposal was that countries plan data centre and compute capacity the way they plan energy.

  3. Regional blocs are building their own financial links, anchored in shared history and trade agreements.

    Cooperation among the Association of Southeast Asian Nations (ASEAN) plus three was traced to the Asian financial crisis, and Africa's continental free trade agreement, through its digital trade protocol, now lets domestic payment schemes start connecting. Schemes signed a memorandum in Penang on ASEAN-level payments beyond bilateral links. In one case, a country with its own currency moved settlement onto European infrastructure, keeping monetary sovereignty separate.

    One proposal was common interoperability standards, so markets on different ledgers connect while keeping sovereignty over their own rails. Against that, one session heard that geopolitics makes allies and passporting more necessary just as trust between countries is at a low.

Payments & Cross-Border Connectivity

  1. Bilateral payment links deliver progress now, but speakers warned they turn heavy beyond a handful of connections.

    In Europe's experience, bilateral links between clearing houses became operationally heavy after 5 or 6 connections, adding liquidity strain, so lower-volume channels were left unconnected. Central banks of five Association of Southeast Asian Nations (ASEAN) countries and India are building a hub-and-spoke gateway, Project Nexus, targeting a live launch next year. Ghana's next link will be with Nigeria, following the trade corridor.

    One speaker argued the technology is the simplest part. The harder work, they said, lies in operating rules, contractual certainty, a liability regime and multilateral supervision, which decide whether linked corridors carry volume.

  2. Cross-border links are being built as connectors between national systems rather than replacements for them.

    One panel argued that links should respect each country's payment sovereignty and local standards. Over 60 foreign wallets across 16 markets can now read a domestic Chinese network's quick response (QR) codes directly, building on work dating back to 2018, and Japan's and Singapore's QR schemes were connected live on stage.

    The convergence has a flip side: rails into China vary, but in one case they converge on a single merchant network. One address argued that as AI agents route payments, the cost of switching to alternative rails will fall sharply, which could loosen such concentration.

  3. Asia Pacific sits on most major trade corridors, and crossing borders still multiplies payment costs.

    The stage heard that 70 of the world's top 80 trading corridors include an Asia Pacific market, and flows between China and the rest of the world were picked as the corridor to watch. One session's own research found a payment from Europe to outside the Single Euro Payments Area costs 12 times as much, and migrants sending money home from Europe pay about 6% on average.

    For providers, speakers pointed two ways: one argued they must prepare for business-to-business flows, put at 70% of cross-border flow by 2030, while another argued stablecoins can make remittances significantly cheaper.

Inclusion, Sustainability & Emerging Markets

  1. National public rails brought millions of people and merchants into view of the financial system in India and Brazil.

    In India, digital public infrastructure, including a national digital identity and simplified customer checks permitted by regulation, helped bring about 560 million inclusion customers into banking, roughly a third of the population. Brazil's instant payment system, launched in 2020, multiplied per capita digital payments 3.6 times in five years and made 22 million merchants visible to lenders.

    One speaker argued that counting transactions is the wrong metric for public systems, and that operators should measure who the transactions reach and how much they change users' lives.

  2. A country's payment model exports its principles, not the market conditions that made it work.

    One speaker argued that what transfers from India's system is interoperability by design, common standards, a level playing field and a simple, affordable experience, while population size, banking depth and identity infrastructure do not.

    Speakers framed this as a gap between the right intent from governments and regulators and effective implementation on the ground, which matters for any country planning links between national systems.

  3. Open-source builds gave governments shared infrastructure they could own and afford, but its funding model split speakers.

    In one country, a local team built a national instant payment switch on an open-source platform between 2020 and 2024, and it now handles 1.5 million transactions a day where no switch existed. In another case, an open-source government mail system, chosen because a proprietary platform was unaffordable, proved stable after five years.

    One speaker held that government ownership is critical and philanthropic capital should work with it; another argued public and philanthropic money should be catalytic and then withdrawn, leaving private innovators to sustain the platform. A third view favoured not-for-loss rails with modest fees reinvested in security and growth.

Sessions reported AI agents running at scale in lending, compliance and security work, with measured cuts in time and errors. Agent-initiated payments are working in tests and first live transactions, but remain a small share, while fraud and quantum risk press on controls.

AI & Agentic Finance

  1. Agents in production are cutting work from hours or weeks to minutes or days in reported cases.

    In one case, a security operations centre run by a first-tier agent cut triage on about 2,000 cases a month from about 15 hours to about 20 minutes, at around 98% accuracy. In another, agents preparing data cut customer onboarding from about two and a half weeks to about two days.

    Breadth still lags depth. One session put AI use at around 80% of banks but enterprise-level use at 10%, and a UK survey cited on stage found only 2% of firms deploying agents with broad autonomy. One speaker added that bank cost-to-income ratios have not yet moved.

  2. Agent-initiated payments work in tests and first live transactions but remain a sliver of activity.

    In one case, about 15% of Asia Pacific transactions are agent-influenced while fully agentic purchases are under 1%, and another speaker said only about 2 to 3% of agentic tasks end in a payment. Checkout design decides success: 98% of trusted agents completed checkout on forms carrying machine-readable instructions, against 50% on forms without them.

    Speakers described most agent transactions today as small-value tests. In one case, conversion on a rushed agent checkout fell below 50% within months and forced a pull-back, and a demonstrated treasury build ran in a test environment with no real money moved.

  3. Unit prices for AI are falling fast, but speakers warned total bills grow as usage climbs.

    One session put the fall in AI service prices at roughly 280 times over three years, and another reported price gaps of up to 100 times between models doing similar tasks. Proposed controls included tiered model routing, caching and budgets on the agent's reasoning loop.

    The outlook split. Falling unit prices are driving heavier use, so bills are rising, and one view held that firms now pay heavily subsidised prices that will bite within three to five years. Speakers also split on whether models are interchangeable, one view saying they are converging and another that swaps may not give similar results.

Risk, Fraud & Security

  1. Sessions converged on verifying the human behind each agent and checking its authority on every transaction.

    Speakers described know-your-agent (KYA) checks being added alongside business and transaction checks, so the person behind each agent can be verified and held responsible. In one case, a protocol lets merchants recognise an agent that is who it claims to be and acts with customer consent, instead of admitting or blocking every bot.

    Speakers differed on where this layer should sit. One view held that KYA checks need continuous monitoring shared across the parties to a transaction. Another held that regulators, payment operators and card networks are each building verification separately, leaving an independent layer across them missing.

  2. Fraud is getting faster and harder to spot, while AI defences post measured gains in reported cases.

    One session said AI tools have cut the time to turn a vulnerability into a working scam from days or weeks to minutes, and a report cited on stage found the share of easy-to-detect fraud in Singapore fell from 95% to 42%. On defence, AI monitoring of about 80 million events a day cut fraud losses by 20% in one case.

    Detection is moving across institutions: a proof of value has five banks pooling transaction data so AI models can spot scam patterns a single bank cannot. One session argued that fixed rules reviewed each quarter cannot keep pace and that controls should adapt to each customer.

  3. Resilience is resurfacing as a design limit, from cloud outages to recovering services if AI fails.

    One session recounted that when cloud data centres in the United Arab Emirates were hit during geopolitical tensions, two large banks on decades-old systems kept operating while fintechs relying on those centres went down. Another heard a commitment to keep enough people, and independence from any single model supplier, for services to be recovered if AI fails.

    Speakers argued that agents raise the bar because they work without pause, so their data must stay available through regional outages. One session claimed zero downtime and zero data loss from consensus replication across nodes, offered as a guarantee rather than a measured result.

Payments & Cross-Border Connectivity

  1. Quick response (QR) code interoperability is delivering growth, and one-tap deep links cut friction where schemes support them.

    One session reported QR payments in Malaysia growing about 60% a year for five years and overtaking cards by value and volume in 2025, credited to interoperability. In one case, a network links over 60 foreign wallets across 16 markets; in another, about 200 wallets in 37 countries and regions. In one marketplace, deep linking roughly halved payment completion time and lifted success rates by close to 9%.

    The limits were stated on stage: deep links work only where banks, wallets and schemes support them, and QR volumes may still trail card payments.

  2. Cross-border speed is largely there on some networks; reliability and governance are now the harder problems.

    On one network, half of cross-border payments reach the beneficiary bank in under 40 seconds and three quarters in under 10 minutes. In one case, a provider serving about 70 banks converts conventional interbank messages into real-time payments, letting banks go live without changing core systems, though not in every corridor.

    Research presented in one session found customers care more about a guaranteed payment than real-time speed, and failed payments cause the most strain between partners, prompting redundancy and intelligent routing. Experiments with shared ledgers, synchronisation and hub-and-spoke models all worked, leaving governance and common rules as the harder task.

  3. National instant payment systems now run at a scale where resilience and fraud controls are design essentials.

    One session reported a national instant payment system handling around 87 million transactions a day, making resilience and fraud prevention matters of public trust. In another country, the instant scheme is expected next year to reach shop checkouts by phone tap, an alternative if international card rails fail.

    One speaker said in hindsight that fraud prevention should have been built in from launch, as many users lost money to scams. Another said a link between cards and instant rails that lets users keep card rewards while merchants receive instant funds, including across borders, does not yet exist.

Digital Assets, Stablecoins & Tokenization

  1. Tokenised settlement is running at volume, with shared ledgers live across banks and currencies.

    One session reported a shared ledger for interoperating tokenised deposits across five major currencies, live since July and used by 19 banks. Another reported a blockchain account-to-account settlement network linking 23 banks that processed 1.1 trillion US dollars in the last 12 months, with 2 trillion forecast for next year.

    Scale now rests on standards. One proposal estimated that a common framework for tokenised bank liabilities could cut cross-border transaction costs by 12.5%, and another speaker argued that scaling needs tokenised central bank money and protocol standards for interoperability.

  2. Programmable money is being shaped for agent payments too small for card fees to handle.

    One speaker put the average agent payment at about 30 cents, too small for card fees, and argued for digital dollars on blockchain rails, with work under way on the x402 standard for machine-to-machine micro-transactions. Controlled trials are also set out in which AI agents use purpose-bound money that pays automatically once embedded conditions are met and checked.

    Speakers attached conditions. Faster machine-to-machine payments hold only if security, identity and governance keep pace, and one proposal held that the institution applying an embedded rule should stay liable even when the rule is wrong.

Quantum

  1. Both quantum sessions argued for crypto agility rather than a one-off cut-over to post-quantum algorithms.

    Speakers warned that attackers can collect encrypted data now and decrypt it once quantum power suffices, which ties exposure to how long data is retained. Both sessions recommended crypto agility, the ability to switch, test, measure and roll back algorithms and keys if today's post-quantum algorithms are later broken.

    Readiness is uneven. Observations cited on stage put about 70% of traffic at the network edge as quantum ready, with only about 15% behind the edge running hybrid post-quantum cryptography. One session argued readiness should be judged by closing classical channels rather than by meeting timeline milestones.

  2. Post-quantum signing is markedly slower, so system owners must test latency and throughput at payment volumes.

    A test cited on stage found post-quantum signing of message batches took about 210 milliseconds against about 28 milliseconds for classical cryptography. Speakers said migration belongs with system owners as well as security teams, since they must weigh latency and throughput penalties on platforms such as payment rails.

    Vendor dependence is the other exposure. In a survey of banks cited on stage, most relied on vendors for quantum safety but only 8% had a contingency plan. One session urged a fallback for any milestone resting on a vendor, such as switching supplier or re-architecting legacy systems.

  3. Quantum computing in finance is at small-scale experiments that do not yet beat classical methods.

    One session placed the field at small-scale demonstrations of working principles, with error-correction engineering under way, alongside a nationwide pilot of quantum key distribution to secure communication channels. Research showed a potentially large speed-up for portfolio optimisation.

    The gains are conditional. Extracting the full allocation across every asset loses that speed-up, while a single figure such as portfolio risk may preserve it. A claim that quantum methods could raise fraud detection rates came with no loss figures.

Sessions held that firms stay accountable for what AI agents do, with runtime safeguards now sorting each agent action. Licence passporting, settlement in central bank money and tokenisation law moved forward, while speakers named governance rather than technology as the cross-border constraint.

AI & Agentic Finance

  1. Speakers held that deploying AI agents leaves firms accountable, so the duty falls on control design.

    Sessions described accountability staying with the institution: agents act within defined authority and limits, with safeguards built in from the start. In one case, agents will not be allowed to override risk, financial crime or sanctions controls, because pointing to an agent is no defence. The stage heard repeatedly that tasks can be delegated but responsibility cannot.

    Speakers differed on scope. One view held that firms should answer for the systems of control around agents rather than for every agent action, while another placed accountability with whoever provides agent verification to the merchant, noting the rules behind those protocols are still being defined.

  2. Singapore now has a runtime safeguards framework that sorts each agent action into one of four outcomes.

    Speakers described the framework, developed with industry and shipped with reference code, as giving each agent action one of four outcomes: execute, execute while flagged for monitoring, escalate to a human, or reject, each logged in a tamper-proof audit trail. Agent guidance from the region was published in July, and India's retail payments operator signalled in September how agents will be identified on its real-time payment system.

    The design answers a scale problem: one speaker argued human approval of every agent action cannot work at machine speed. In one demonstration, a deterministic rules engine denied a purchase that breached the client's mandate, with exceptions escalated to a human treasurer.

  3. Existing rulebooks already reach AI in finance, from EU credit-scoring duties to doubts over the three lines of defence.

    One session set out the EU AI Act treating creditworthiness checks as high-risk, with duties on both providers and deployers and record traceability adding audit and storage cost. Under Indian rules, one pitch explained, model retraining must pass a separate validation process before going live. One speaker said the question in the UK is whether the traditional three lines of defence still fit as teams of agents make decisions harder to trace.

    One speaker argued that AI model developers do not take liability when things go wrong, so regulated payment firms must supply the liability and dispute framework, as card networks did before online card payments took off.

Policy, Regulation & Central Banking

  1. Licence passporting is moving from talk to pilots, with Ghana and Rwanda opening to eligible fintechs from December.

    The Ghana and Rwanda arrangement has piloted 2 entities at each end and proved less difficult than expected, speakers said. Passporting already works in the West African monetary union, where national offices recognise regional central bank approval on notification, and Europe's single market lets one licence reach 450 million consumers. Singapore has consulted on recognising stablecoins regulated elsewhere, issued against a shared reserve pool.

    Speakers also drew limits. One view held that a host supervisor must keep its right to intervene alone in an emergency, such as illiquidity, and another argued that since rules will not be identical, markets should first agree interoperable principles.

  2. Tests showed clearly defined compliance checks can run inside the payment rails, while checks needing judgement cannot.

    One session reported that limits and on-chain verification of compliance proofs proved implementable, while discretionary checks could not be fully automated. In one project test, an outdated sanctions list ran on chain and the system raised no alert, prompting calls for rule ownership, override powers and an audit trail. One speaker called for a kill switch for regulated banks, and for settlement finality in atomic settlement to be defined in law.

    The stage heard that code cannot be held liable, as deposit insurance covers a bank failing but not code failing. One speaker added that on-chain compliance remains experimental and that freezing funds is hard to reconcile with atomic settlement.

  3. Settlement in central bank money is live in the euro area, with a UK synchronisation service due in 2028.

    Speakers reported tokenised transactions settling in central bank money in full production in the euro area since the end of September, with large issuances. In the UK, a service synchronising market platforms with central bank settlement, so both legs settle together, has 18 firms testing it. In one jurisdiction's experiments, synchronisation proved a fast, low-cost route that avoids building a wholesale central bank digital currency platform.

    Retail is less settled. One speaker expected a digital euro by 2029, while a view from a small currency area saw no strong case yet for a retail central bank digital currency, though offline capability could change that.

Digital Assets, Stablecoins & Tokenization

  1. Regulated stablecoins still meet legal blockers because prudential and accounting rules do not treat them as cash.

    One speaker argued that regulated stablecoins should count as cash equivalents across accounting, fund, clearing and prudential rules, since they are not treated as cash or high-quality liquid assets and each use case meets a legal blocker. In one case, a subsidiary has issued euro and US dollar stablecoins under EU crypto-asset rules for two years. One session set out safeguards common to three regimes: reserves held one to one, a bankruptcy-remote structure and monthly third-party attestation.

    Rules alone do not create use. One speaker argued Europe already has a reasonable stablecoin rulebook but lacks demand and the right infrastructure.

  2. Thailand is amending fund and securities law so tokens carry legal force, while a security's legal nature stays unchanged.

    One panel agreed that tokenisation changes how a security is held, traded and transferred, not its legal character. Thailand amended mutual fund rules in April to allow tokenised fund units with real-time settlement, and the first tokenised money market fund was offered to the public in September. Its securities law is being amended so security tokens are enforceable across their life cycle, notably as collateral.

    Protection depends on labels. The stage heard that tokenised deposits remain insured bank deposits, whereas a deposit token, where the token itself is the record and the holder may be unknown, is not a deposit and not insured.

  3. Speakers argued tokenised markets need central bank money to scale, with commercial bank money tokenised alongside it.

    One panel argued that tokenisation needs central bank money in the mix to reach large scale, treated as a public-private partnership in which public solutions drive private innovation. Another view held that tokenised central bank money is not enough on its own, and that banks must also tokenise commercial bank money as deposit tokens, alongside euro stablecoins for specific uses. In France, an entire commercial paper market is due to settle against tokenised central bank money, going live in December if things go well.

    Another panel argued that policymakers focus too much on issuance, and that stablecoins will scale only if they are widely accepted and redeemed at par.

Payments & Cross-Border Connectivity

  1. Cross-border settlement experiments worked technically, leaving governance and common rules as the binding constraint.

    Speakers reported that shared ledgers, synchronisation and a hub-and-spoke model all worked and delivered the same functions, and that real-value testing with around 50 commercial banks settled in seconds payments that usually take days. A multi-central-bank proof of concept showed atomic wholesale settlement can work while preserving currency and operational control.

    What remains is the rules. One speaker argued linking instant payment systems needs a shift from bespoke bilateral links to a standardised multilateral arrangement with common scheme rules, governance, liquidity and compliance. In Europe, a cross-currency instant payment solution is stuck between the technology, the payment scheme and European regulation.

  2. Mandated open access is spreading payment links, but one session warned that pricing set too low drives operators away.

    One speaker reported being required by Malaysia's central bank to open their merchant network to every issuer, wallet and bank, free of charge, by joining the national quick response (QR) code scheme. Cross-border QR and account-to-account links between the region's national schemes are in real use and growing, the same session heard.

    The counterweight was commercial. One speaker argued regulators must design the commercial model as well as the rules, because unsustainably low pricing drives private operators away, and that connectivity scales only with common technical standards, clear accountability between schemes and incentives that work for every party.

Sessions described money in Southeast Asia shifting to growth stages, leaving a Series B gap, while speakers placed lasting value in data, networks and licences rather than AI models. Small-firm lending on cash-flow data showed early scale, and institutional stablecoin use waits on capital treatment.

Investors, Entrepreneurship & Scaling

  1. Regional capital is concentrating in growth and late stages, leaving a Series B gap that raises bars earlier.

    In one session on Southeast Asia, speakers reported that funds have returned less capital than they deployed, so more money goes to growth companies close to an exit. Globally focused investors have cut regional deal activity and moved towards AI elsewhere. A large gap at Series B pushes up Series A expectations, so pre-revenue companies are very hard to fund.

    Speakers differed on the cause. One view held that the slowdown is both structural and cyclical and not driven by interest rates, while another held that 2021, not today, was the aberration. Another session found that startups surviving the pivot point tend to become revenue funded, not investor funded alone.

  2. With AI models open to all, speakers placed lasting value in proprietary data, networks and licences.

    Several sessions argued that access to a capable model no longer sets a firm apart and that AI savings will be competed away. In one case, a credit limit increase pilot combining card, deposit and wealth data showed a 10 to 15% lift in net present value per customer granted an increase.

    Speakers split on what still defends a business. One view discounted moat claims in early-stage investment memos and called AI a moat killer. Others held that network effects remain strongest, because trust and liquidity take time to build, and that a limited number of licences can be an effective moat.

  3. Bank relationships carry concentration risk for young firms, and one proposal had banks fund fintech-made loans.

    In one case, a firm's only bank ended the relationship at about one year with no discussion. The agreement required 30 days' notice, but the bank agreed to six months when asked, and the firm has since gone from one counterparty relationship to more than 15, with its own custody relationships. Elsewhere, over 60 bank collaborations for venture-backed portfolio companies have been arranged since 2019.

    One proposal had commercial banks, for which very small loans cost too much to originate, invest in portfolios that fintechs originate while fintechs keep underwriting. The counter-case heard on stage is that incumbents' customer bases and client trust remain the real moat.

Inclusion, Sustainability & Emerging Markets

  1. Cash-flow data is starting to stand in for land as collateral in lending to small firms.

    One session put the small-firm financing gap at 6.3 trillion dollars, with about 60% of small firms in emerging markets invisible to lenders because they do not use digital payments. Analysis presented there found that scoring on five years of revenue and expense data predicts creditworthiness better than land. In one case, cash-flow lending on account, tax and bureau data underwrote around ₹1 trillion last financial year, 20% of its small-business loan book.

    Pressed on whether lenders actually lend on data alone, the same session conceded the solutions remain incipient, so the evidence is early rather than settled.

  2. Development finance is carrying part of the credit risk so that commercial lenders reach small firms.

    In fragile and conflict-affected markets without data trails, one session described risk-sharing facilities in which a development lender carries 50% of the credit risk on targeted small-firm loans and the commercial bank the rest. In another case, a $500 million collateral risk facility uses AI and data insights to push bank issuers to lend to small firms in over 100 developing markets.

    The public side is also working on the data layer. Under a small-business financing mission, AI firms will be invited to build and test credit assessment tools on realistic synthetic data, with lenders and public agencies contributing data.

  3. Lenders using everyday transaction and loyalty data report reaching borrowers that credit bureaus do not see.

    In one case, a lender built on mobile, behavioural and engagement data served 14 million customers and delivered 9 billion in access to credit over about ten years, and is now opening that stack to other fintechs, retailers and telcos. In another, a points programme with 54 million members became a lending business scoring invisible people. In a third, an AI risk platform turns away about 40% fewer good customers by mistake.

    Speakers split on where this leads. One view held that AI could let small firms compete with large ones, another that AI will widen the gap between rich and poor, making trusted data-based credit tools more urgent.

Digital Assets, Stablecoins & Tokenization

  1. Prudential and accounting treatment, not technology, holds back institutional use of stablecoins as cash.

    One session heard that regulated stablecoins are not counted as cash or as high-quality liquid assets, so each use case meets a legal blocker. It also heard a proposal to treat them as cash equivalents across accounting, fund, clearing and prudential rules. Even so, speakers reported European market infrastructures adopting stablecoins as the cash leg of tokenised repurchase agreements.

    Speakers split on the settlement asset. One view held that deposit tokens suit institutions more comfortable with commercial bank money. Another held that the market will decide by jurisdiction, with stablecoins winning in some and tokenised deposits in others.

  2. Stablecoin issuers are competing on distribution, in one case passing reserve returns to partners holding customers.

    In one case, an issuer passes the return on its reserves to the wallets and payment firms that hold the customers, having concluded it could not win distribution alone. A bank consortium with a combined balance sheet of €15 trillion and half of European deposits expects to launch a euro stablecoin in December or January, with member banks connecting their wallets.

    Demand is the open question. In one case, most euro stablecoin assets come from retail users, with demand now shifting towards market infrastructures, while another session saw corporate demand in managing liquidity and collateral.

  3. Banks are beginning to make markets in stablecoins, though few institutions yet transact on public blockchains.

    In one case, a bank now acts as liquidity provider for its stablecoin on a crypto exchange, part of a wider move by banks into on-chain market making and foreign exchange liquidity. Another stablecoin is live as a trading and collateral asset at three large crypto venues and in on-chain lending liquidity.

    Yet only a small handful of financial institutions transact on public Ethereum today. One speaker expected the industry to settle on a small number of open networks, perhaps three to five or five to 10, within a couple of years, rather than one network or one per bank.

Sessions reported falling employment among young workers in AI-exposed roles, while in one case freed capacity was redeployed rather than cut. Retraining, how firms reorganise around AI agents and how founders survive and hire also featured strongly.

Macro, Geoeconomics & Talent

  1. Early displacement of young workers in AI-exposed roles is now measurable, though speakers split on whether it lasts.

    One address reported that in the US, employment of 20 to 25 year olds in the most AI-exposed occupations has fallen about 13% relative to their peers since generative AI took off. Another session described early displacement of young workers in professional services and finance, arguing banks that trained recruits on tasks AI now does need another way in.

    One view expected almost every job to change without a long shortage of work. Another, citing headcount already falling in one case, urged policymakers to prepare for higher unemployment. In a separate case, AI agents handle 75% of customer service tickets and freed capacity is redeployed, not cut.

  2. Speakers put retraining at the centre of the AI transition, some arguing industry must carry much of it.

    One session committed to protecting every worker, if not every job, through retraining for roles AI creates. Another reported that 10 to 12 million young people join India's workforce each year, making reskilling the top priority, and argued industry must lead it. A third said banks should give front-line staff AI skills and AI developers business knowledge.

    One session argued that oversight adds to the bill: human checks on high-risk AI mean staff must be trained to be AI literate. One view held AI literacy will soon stop counting as a technical skill, as computer literacy did, which would make it a baseline expectation rather than a specialist hire.

  3. In cases described, organisations deploying agents are being rebuilt so people define problems and own outcomes.

    In one case, the organisation behind an agent was rebuilt so it can call on the whole bank with a single command, with people shifting to defining problems, making judgements and taking responsibility. In another, technology teams long held in a separate legal entity are being moved into the same room as the business. A third case reported 97% of staff using AI daily.

    One view held that autonomy can pass to agents while accountability stays with people, so reskilling must be continuous. The strongest programmes were described as pairing bottom-up building by staff with top-down standards and joint sign-off across business, risk, compliance and finance.

Investors, Entrepreneurship & Scaling

  1. One analysis found only 10% of startups survive a pivot point that founders reach four to five years in.

    The session's pattern analysis found defining the problem typically takes six months to two years, and the stretch after market and investor feedback is often a cliff rather than a slope. About half of those whose startups ended were judged right to close. It also said 80 to 90 founders in its community are struggling now behind public announcements.

    The other half most likely needed to persist, and the session held founders find it hard to know at the time which choice is right. Another address argued the test is admitting a wrong turn and changing course, rather than pressing on because of what has already been invested.

  2. India's founder base is young and widening, with students from smaller cities starting many new firms.

    One session reported registered startups in India rose from 500 in 2016 to close to 300,000, many founded by students from Tier 2 and Tier 3 cities. The average age of the country's unicorn founders was given as 33, and often it is not their first company.

    Another session, drawing on a failed startup, painted a different picture of Singapore: with risk-taking still uncommon, early hires are hard to attract, so founding teams must carry a company to product-market fit themselves. A wider pool of founders does not by itself settle who will join a startup before it works.

Inclusion, Sustainability & Emerging Markets

  1. Formalisation of India's informal workforce was described as under way, but without firm data behind it.

    One session said the informal sector, once estimated at half of India's economy, is shrinking because of social security for informal workers, small-enterprise registration and digital public rails, while noting no firm data could be quoted. Another reported a skills gap among users: more than 80% of adults in India hold a bank account, but financial literacy stands at 24% and adviser access below 5%.

    That session's team proposed an AI assistant giving guidance in native languages and budgeting around irregular, seasonal incomes. Both point to the gap moving from access to capability, though the formalisation trend remains a direction of travel rather than a measured result.

Day 2 at SFF

What is on the agenda for Day 2.

Covers cross-border corridors: Asia-Middle East and Gulf-Asia-Africa connectivity, live tokenised corridors, the cost of linking national payment systems, and treasury and small business trade across Greater China and Southeast Asia. Also turns to which Asian markets gain or lose ground as integration grows more contested.

The agenda centres on agentic AI: agent-initiated payments and commerce, whether payment rails and bank architecture can run at machine speed, and moving AI from pilots to production. Tokenisation plumbing, stablecoins, tokenised deposits, interoperable rails and passkey authentication fill out the day.

Covers accountability when AI agents act, including who regulators supervise when agents direct other agents, the compliance perimeter in core banking and compliance written into code. Also trusted hubs for programmable money, cross-border payment cooperation, identity rules on-chain and protection for first-time investors.

Covers which money institutions should hold - stablecoins, tokenised deposits or tokenised money market funds - and balance-sheet exposure in programmable money. Also founder exits and scaling, investors in tokenised real-world assets, financing infrastructure in emerging markets, small business credit and first-time retail share ownership.

Covers how agentic AI changes jobs and organisation design, scaling human and AI collaboration inside banks, and what growing companies should automate or still hire for. Talent pipelines feature through a fintech talent report, a young talent programme for AI and a polytechnic hackathon.