Most discussions about artificial intelligence begin with the same question: which jobs will disappear? That framing misses a more significant shift. The real economic impact of AI may have less to do with intelligence and more to do with speed. For most of modern commerce, decision-making and payment have been separate events. A customer searched, compared options, made a choice and then initiated payment. Even online, transactions continued to move at human speed. A person remained the final checkpoint.
Agentic AI changes that model. Software can now be given authority to search, negotiate, select suppliers and complete payments within predefined parameters. The gap between decision and settlement begins to disappear. This is not simply another automation story. It is a fundamental change in how commerce operates. When money can move at machine speed, entirely new forms of economic activity become possible. The scale of the opportunity is already becoming apparent. Accenture estimates that more than 30% of online commerce could be routed through AI agents by 2030, representing approximately $3.1 trillion of activity. J.P. Morgan has highlighted industry forecasts suggesting that AI agents could account for $2-5 trillion of commerce by 2030, representing approximately 3-7% of global payment volumes. Whether those figures ultimately prove optimistic is almost beside the point. They illustrate a broader reality: software is moving from supporting transactions to executing them. The debate, therefore, should not focus solely on whether AI removes jobs. The more interesting question is what new forms of work become necessary when software can both decide and pay.
The moment payments become autonomous
Until recently, digital tools could recommend products, compare suppliers or optimise procurement decisions. Humans still had to press the button. Increasingly, that button is disappearing. Industry expectations suggest that adoption will begin where risk is lowest and commercial benefits are most obvious. J.P. Morgan anticipates agentic purchasing first becoming mainstream in repeatable consumer purchases such as groceries, travel and ticketing before moving into more complex categories. In the corporate world, procurement is an obvious starting point. Many organisations already operate using approved supplier lists, spending limits and escalation protocols. AI agents simply automate decisions that already follow established rules. Machine-to-machine commerce is likely to develop even faster. Mastercard’s Agent Pay for Machines initiative, for example, is designed to facilitate autonomous low-value transactions between systems, including purchases of data, storage and computing resources. Those transactions often occur at a frequency and scale that would be impossible to execute manually. The significance is not simply that payments become faster. It is that entirely new markets become viable when the cost of making a transaction approaches zero.
The real constraint is not technology: it is authority
The most common assumption is that the challenge will be technical. It is not. Identity solutions will improve. Payment rails will become more sophisticated. Authentication methods will evolve. The more difficult question is who authorised the transaction. Traditional commerce is built around clear chains of authority. Agentic commerce introduces a new participant: software operating under delegated authority. That distinction matters because authority and accountability are not the same thing. If an AI agent exceeds its mandate, purchases from a fraudulent supplier, misunderstands an instruction or causes financial loss, organisations will need to answer difficult questions. Who authorised the agent? What restrictions applied? Who was expected to monitor its activity? Who bears responsibility? Industry concerns suggest these questions are already front of mind. A recent Cloud Security Alliance survey found that 85% of respondents expect AI agents eventually to initiate consumer payments, whilst 65% believe doing so will require entirely new authorisation frameworks. These are not technology questions. They are governance questions. In many ways, the next phase of digital commerce will be defined less by advances in AI capability and more by advances in organisational trust.
Trust has always lagged innovation
History provides a useful guide. Online banking did not succeed because consumers suddenly trusted the internet. It succeeded because banks, regulators and technology providers-built frameworks that allowed trust to catch up with convenience. The UK’s experience illustrates the point. By 2022, 88% of UK adults were using online banking or mobile banking services. Yet concerns around fraud, scams and digital security remained persistent throughout the sector’s development. Agentic payments will require a similar evolution, but arguably on a larger scale. An AI agent may be acting on instructions given weeks earlier. It may be interacting with another AI agent rather than directly with a human. It may complete thousands of transactions before anyone reviews its activity. As commerce accelerates, trust becomes a competitive advantage. The organisations that succeed will not necessarily be those that automate most aggressively. They will be those that build the strongest governance framework around automation.
Every new payment system creates new disputes
History also suggests that innovation does not eliminate disputes. It changes their shape.
The rise of online banking generated disputes around fraud and payment authorisation. Digital assets created disputes around custody, tracing and ownership. Agentic commerce is likely to follow the same pattern. Future disputes may focus on whether an AI agent acted within its mandate, whether a merchant was entitled to rely on instructions provided by software, or where liability sits when multiple autonomous systems interact with one another. That should not be viewed as an obstacle to adoption. Every step-change in financial services has required markets to develop new mechanisms for allocating risk. The important point is that governance can no longer be treated as an afterthought.
The jobs that slowness made impossible
This is where the employment debate often goes wrong. The future is unlikely to be defined by a straightforward replacement of people with machines. Instead, value shifts towards activities that become more important as commercial activity accelerates. Evidence of that shift is already emerging. LinkedIn job-posting data reported by the Wall Street Journal suggests that approximately 640,000 AI-related roles were added in the United States between 2023 and 2025, reflecting growing demand for people who can design, deploy and govern AI systems. Separately, in a recent survey of 100 fintech leaders commissioned by Chatsworth found that 90% expect to increase UK investment and hiring over the next three years. Those statistics do not prove that AI creates more jobs than it eliminates. What they do suggest is that entirely new functions emerge whenever new economic infrastructure is built. As organisations deploy autonomous payment systems, demand will grow for specialists who can design governance frameworks, supervise AI decision-making, verify digital identities, oversee compliance controls and investigate failures.
The critical skill is not making the payment. The critical skill is determining when the payment should be made, under what authority and with what safeguards.
Faster money only works if it still settles
Much of the discussion around AI payments focuses on speed. Speed is only half the story. The real prize is efficiency. Barclays and PwC have estimated that tokenisation could contribute up to £33 billion annually to UK economic output by 2035 through lower transaction costs, improved capital efficiency and faster settlement. If AI agents can identify suppliers, source financing, manage treasury activity and complete transactions in near real time, organisations will be able to deploy capital more effectively and reduce friction across global operations. But none of those benefits matter if participants lose confidence in the system. Commerce does not run on technology alone. It runs on trust.
A new participant in commerce
The biggest misconception about agentic commerce is that it removes humans from the transaction. In reality, it changes where humans add value. As software takes on more of the mechanics of buying and paying, value shifts towards defining authority, managing risk, resolving exceptions and maintaining trust. The winners of the next phase of digital commerce are unlikely to be the organisations with the most autonomous agents. They will be the organisations that understand how to govern them. Commerce is acquiring a new participant: software with delegated spending power. The businesses preparing for that future today are not asking whether AI can make payments. They are asking a far more important question.
Are our governance, controls and accountability frameworks ready for a world where it can?
This article first appeared in Digital Bytes (20th of October, 2026), a weekly newsletter by Jonny Fry of Team Blockchain.
How will AI agents change the way we buy, sell and pay?