When Visa announced it would cut about 2,600 jobs, or 7 per cent of its workforce, it was easy to read the decision as another sign that artificial intelligence has begun replacing white-collar workers. Visa is hardly alone. Mastercard has cut about 4 per cent of its global workforce, though it attributed the decision to a strategic review rather than to AI.
But we should be careful about drawing too straight a line from AI to layoffs.
Visa announced the cuts in the same week it reported its strongest quarter on record, with net revenue up 14 per cent, quarterly payments volume crossing US$4 trillion (S$5.1 trillion) for the first time, and full-year guidance raised. This is not a business being displaced. It is one reallocating capital from strength.
Companies cut headcount for many reasons: slower growth, past overhiring or strategic change. There is also an incentive to describe restructuring as an AI transformation, because a leaner, AI-enabled organisation is a better story for investors than a company that overhired. When Mastercard made its cuts, it reached for none of that language, a reminder that invoking AI is a choice rather than a description.
The broader evidence also argues against declaring an AI jobs apocalypse. If AI were already replacing labour at scale, we should expect stronger employment effects and visible productivity acceleration. Neither pattern is overwhelming yet. Research finds little systematic rise in unemployment among the occupations most exposed to AI, though Stanford economists see early signs that entry-level hiring is weakening.
The layoffs therefore do not prove that AI has displaced thousands of workers, but neither should we conclude that little is happening. The more important transformation may occur inside jobs rather than in the unemployment rate.
Visa is in fact facing two technological shocks. AI is a general-purpose technology reorganising work across knowledge-intensive industries. Tokenisation, including stablecoins and tokenised bank deposits, is specific to payments, because it changes the infrastructure through which money moves.
Put simply, AI threatens some of the tasks performed inside Visa. Tokenisation challenges parts of what Visa itself does.
Which tasks, not which jobs
The debate is usually framed as a choice: AI either substitutes for humans and destroys jobs, or complements them and raises productivity. Both can happen inside one occupation. The better question is which parts of a job can be delegated, and what becomes more valuable once they are.
Consider a call centre. Much of the work follows a recognisable sequence: understand a question, retrieve information, apply a policy, communicate the solution. An AI agent can increasingly execute that sequence itself, and where tasks are repetitive, rule-governed and cheap to verify, the case for a person performing every step weakens rapidly.
An auditor investigating an unusual transaction is different. AI may read thousands of invoices in seconds and offer several plausible explanations, but somebody must still decide whether an anomaly is error or fraud, and whether the institution should act. That final step requires domain knowledge, professional scepticism, context and responsibility. The same technology can therefore substitute for one part of work while raising the value of another.
The first labour-market effect may therefore show not in mass layoffs, but in who companies stop hiring. A bank that once needed 10 junior analysts may find five equipped with AI to do the same preparatory work. Senior staff remain valuable because they can supervise the machine, but fewer juniors come in beneath them.
That deserves attention. Today’s senior auditor learnt by being yesterday’s junior. If AI eliminates the apprenticeship work, where will tomorrow’s experts acquire the judgment that AI cannot provide?
Living with stablecoins
For Visa, AI is only half the story. Tokenisation reaches closer to the economics of the payments business itself.
Payment networks create value by connecting consumers, merchants, banks and institutions across systems that do not naturally communicate. Tokenisation loosens those constraints: stablecoins move value continuously across blockchain networks, and tokenised deposits make bank money programmable.
This does not mean Visa cards are about to disappear. Visa’s own analysis suggests less than 1 per cent of adjusted stablecoin volume – which strips out automated activity – consists of retail transactions below US$250. Most activity sits upstream, in holding, transferring and settling value.
Asking whether consumers will swop credit cards for crypto wallets therefore misses the larger change. The first serious competition comes behind the payment, in cross-border transfers, institutional settlement and treasury management.
Institutional finance is already moving. BlackRock has launched a tokenised money-market fund on public blockchains, and the New York Stock Exchange is developing, subject to regulatory approval, a platform for round-the-clock trading and on-chain settlement of tokenised US equities, funded by stablecoins.
These are not crypto start-ups trying to replace Wall Street. This is Wall Street adopting the technology.
Visa understands this. Rather than betting against stablecoins, it is becoming part of their infrastructure. Its settlement pilot spans nine blockchains and reached a US$7 billion annualised run rate in April, up 50 per cent in a quarter. Its fastest-growing revenue line is no longer moving money but value-added services, up 34 per cent. Threatened by a new network, an incumbent can defend the old one or make itself indispensable to the new. Visa is attempting the latter.
The same logic runs through agentic commerce. As AI agents begin to shop and pay on their owners’ behalf, someone must vouch for the agent. Visa’s Trusted Agent Protocol issues a verified agent identity with an issuer-signed consent record while Mastercard’s Agentic Tokens bind a card credential to a specific agent, merchant and consent policy. Both companies have endorsed Google’s Universal Commerce Protocol and are bidding to become the trust layer for machine-initiated payments. Whether they keep that role, or the protocol owners take it, is the open question.
Impact on Singapore
Singapore’s banks face the same two revolutions. On AI, they should automate predictable work while concentrating human talent where judgment, context and accountability remain essential. On tokenisation, they must ensure that as money moves to programmable infrastructure, banks do not become passive suppliers of deposits behind somebody else’s platform.
If companies begin holding cash in stablecoins rather than bank deposits, banks risk losing part of their traditional funding base. The answer is not to resist the technology, but to offer similar functionality within the banking system. Tokenised deposits remain in regulated banks while allowing faster transfers. DBS is already working with J.P. Morgan’s Kinexys on such infrastructure, helping banks shape the new system rather than connect to it later.
As card issuers, banks must also authenticate agents and set spending mandates when software transacts for a customer. The Monetary Authority of Singapore has signalled it is ready to legislate on stablecoins and proposed capital rules letting banks hold digital assets without punitive charges.
Eventually the two revolutions will meet. AI agents will initiate transactions while tokenised money settles them. That convergence, not any single round of layoffs, is the turning point worth watching. The scarce resource will be the capacity to decide what to delegate, under what constraints, and who is accountable when the machine gets a decision wrong.
For universities including NUS, that is also the challenge. The goal should not simply be to teach students how to use AI, since that will soon be a basic workplace skill. It should be to build enough domain expertise for graduates to know what to delegate, what to verify, when to challenge the machine and, ultimately, when not to trust it.
Singapore can benefit from both transformations if it gets one principle right: automate execution, but invest more, not less, in human judgment. That is the real message behind Visa’s layoffs. The future of finance belongs to neither humans nor machines, but to institutions that know where each should lead.