Imagine a banking hall. It is vast, architecturally distinguished, flooded with golden light from floor-to-ceiling windows. The terminals are polished, the systems humming, the infrastructure immaculate. And it is completely empty. Not a single human being stands at a counter, sits in a waiting area, or speaks to an adviser. The hall is full of activity - transactions are being processed, accounts are being opened, investments are being rebalanced, insurance policies are being negotiated - but every actor in this space is an autonomous agent operating on behalf of a human who is elsewhere, doing something else entirely.
This is not a thought experiment. This is the near future of financial services, and it is arriving faster than most institutions are prepared to acknowledge. The machine customer - an autonomous agent that acts as a buyer, negotiator, or financial decision-maker on behalf of a human principal - is the most significant shift in the relationship between commerce and consumer since the invention of e-commerce itself.
And almost nobody is designing for it.
What is a Machine Customer?
A machine customer is an autonomous agent that participates in economic transactions on behalf of a human or organisation. It is not a chatbot that helps you shop. It is not a recommendation engine that suggests products. It is an entity that evaluates options, makes decisions, negotiates terms, executes purchases, and manages ongoing commercial relationships - all without requiring human intervention at the point of transaction.
The distinction matters enormously. A recommendation engine presents options to a human who decides. A machine customer decides. It has been delegated the authority to act, within parameters set by its principal, and it exercises that authority autonomously. The human is not in the loop at the moment of action. The human set the loop in motion and will assess the results, but the transaction itself is machine-to-machine.
This is the critical shift that separates the machine customer from every previous form of digital commerce. The buyer is not a person using a tool. The buyer is the tool, operating with delegated authority.
The Gartner Prediction
Gartner's prediction that machine customers will represent a multi-trillion-dollar economic force by 2030 is not speculative futurism. It is an extrapolation of capabilities that already exist. Today, algorithmic trading systems execute the majority of transactions on global stock exchanges. Automated procurement systems negotiate and execute supply chain contracts for major corporations. Smart home systems purchase energy, adjust insurance parameters, and reorder consumables without human instruction.
The machine customer is not a future phenomenon. It is a present phenomenon that has not yet been named, recognised, or designed for with the seriousness it demands.
What is changing is the scope, sophistication, and ubiquity of these autonomous economic actors. The machine customer of 2028 will not be a narrow algorithm executing a predefined rule. It will be a general-purpose agent capable of understanding complex financial products, comparing offers across providers, negotiating terms, assessing risk, and making decisions that reflect the nuanced preferences and circumstances of its human principal.
The implications are staggering. Every business that sells to consumers will need to consider that its customer may not be a consumer at all, but an agent acting on a consumer's behalf. Every financial institution that serves retail customers will need to consider that the entity interacting with its systems may be an autonomous agent with delegated authority, not a human with a browser.
Three Categories of Machine Customer
Not all machine customers are alike. I propose a taxonomy of three categories, each with distinct design implications for the organisations that must serve them.
The Bound Agent operates within tightly defined parameters. It has been given a specific mandate - "find the cheapest home insurance that meets these minimum coverage requirements" - and it executes within those boundaries. The bound agent is the most common form of machine customer today. It is essentially a sophisticated comparison engine with execution authority. The design challenge with bound agents is parameter specification: ensuring the human principal has expressed their requirements with sufficient precision that the agent's autonomous decisions align with their actual preferences.
The Adaptive Agent operates within broader parameters and adjusts its behaviour based on accumulated experience. It learns the principal's preferences over time, refines its decision criteria, and may make choices that the principal did not explicitly authorise but that fall within the spirit of the delegation. An adaptive agent managing a household's energy procurement might switch providers, adjust consumption patterns, and negotiate new tariffs based on its evolving understanding of the household's priorities. The design challenge with adaptive agents is trust calibration: the principal must be able to understand, monitor, and adjust the agent's evolving decision-making without micromanaging it.
The Autonomous Agent operates with broad discretionary authority. It has been delegated not just execution but judgment. An autonomous agent managing a high-net-worth individual's financial portfolio might identify investment opportunities, assess risk, rebalance allocations, and execute transactions based on its own analysis - within regulatory constraints and the principal's stated risk appetite, but without seeking approval for individual decisions. The design challenge with autonomous agents is the most profound: accountability architecture. When an autonomous agent makes a decision that results in financial loss, who is responsible? How is the chain of delegation, decision, and consequence made legible?
The Banking Imperative
For financial institutions, the machine customer is not a distant concern - it is an immediate strategic imperative. Banks that cannot serve machine customers will lose them. And because machine customers optimise relentlessly, they will migrate to institutions that offer the best terms, the fastest execution, and the most seamless agent-to-system integration.
This creates a profound inversion of the traditional banking relationship. For decades, banks have relied on friction, complexity, and human inertia to retain customers. Switching banks is tedious. Comparing mortgage products requires expertise. Negotiating better terms requires confidence and time. These barriers to exit have been, frankly, a business model.
Machine customers do not experience friction. They do not feel loyalty. They do not forget to cancel a subscription or fail to notice a fee increase. They are relentless optimisers operating at machine speed.
Machine customers demolish this model. They do not experience friction. They do not feel loyalty. They do not forget to cancel a subscription or fail to notice a fee increase. They are relentless optimisers operating at machine speed. A machine customer will switch a savings account to a competitor offering 0.05% better interest in the time it takes a human to read the notification.
Banks that understand this will compete on genuine value - better products, better terms, better service - rather than on friction and inertia. Banks that do not understand this will discover that their customer base is being systematically optimised away by agents that never sleep, never forget, and never accept a suboptimal deal.
Designing for Non-Human Actors
The design implications of the machine customer are unlike anything the design profession has previously encountered. We are not designing for humans using machines. We are designing for machines acting as economic agents. The entire vocabulary of human-centred design - empathy, emotion, delight, satisfaction - becomes irrelevant at the point of transaction. The machine customer does not experience delight. It experiences optimality.
This does not mean design becomes irrelevant. It means design must operate at two simultaneous levels. At the machine level, the design challenge is protocol: creating systems, APIs, and interaction patterns that allow autonomous agents to discover, evaluate, negotiate, and transact efficiently. This is fundamentally a systems design challenge - designing the infrastructure through which machine-to-machine commerce operates.
At the human level, the design challenge is oversight: creating the experiences through which human principals understand what their agents are doing, assess whether the outcomes align with their intentions, and intervene when they do not. This is the AXD challenge - designing the trust architecture, the observability layer, and the delegation grammar that governs the human-agent relationship.
The two levels are inseparable. A machine customer that operates with perfect efficiency but whose human principal cannot understand or control its behaviour is not well-designed. A machine customer whose human oversight interface is excellent but whose machine-to-machine protocols are inefficient will be outcompeted. The design must work at both altitudes simultaneously.
The Identity Problem
Perhaps the most fundamental challenge the machine customer poses is the identity problem. When an autonomous agent presents itself to a financial institution, who is it? It is not the human principal - it is acting on the principal's behalf, but it is a distinct entity with its own decision-making processes. It is not a simple tool - it exercises judgment, not just execution. It occupies a new category: an entity with delegated identity and delegated authority.
Current regulatory and identity frameworks are not designed for this. Know Your Customer (KYC) regulations assume the customer is a person or a legally constituted entity. An autonomous agent is neither. It is a capability operating under a delegation from a person or entity. The regulatory infrastructure will need to evolve - and it will - but the design infrastructure must evolve first, because the patterns we establish now will shape the regulatory frameworks that follow.
We need a new concept: Know Your Agent. Not just the identity of the principal, but the identity, capabilities, authority scope, and behavioural parameters of the agent acting on the principal's behalf.
I propose a design pattern I call Agent Identity Architecture: a structured approach to establishing, verifying, and managing the identity of autonomous agents in commercial contexts. This architecture must include the agent's principal (who delegated authority), the agent's scope (what it is authorised to do), the agent's constraints (what it is prohibited from doing), and the agent's provenance (who built it, what model it runs, what version it is). This is not just a technical specification - it is a design challenge of the first order, because the way agent identity is presented, verified, and managed will determine whether machine customers are trusted participants in the economy or unaccountable actors operating in a regulatory grey zone.
Experience Without Interface
The machine customer represents the purest expression of what AXD calls the Invisible Layer. The transaction happens without a screen, without a click, without a human present. The experience - from the human principal's perspective - is entirely in the before and after: the delegation that set the agent in motion, and the assessment of the results it produced.
This creates a design challenge that is genuinely new. How do you design an experience that the experiencer never directly experiences? The answer lies in what I call outcome narratives - the designed communication through which an agent reports its activities, decisions, and results to its principal. An outcome narrative is not a transaction log. It is a designed account of autonomous action, structured to build understanding, maintain trust, and enable informed oversight.
A well-designed outcome narrative for a machine customer managing your insurance might read: "I reviewed 47 policies from 12 providers. Your current policy ranked 8th on value. I identified three options that offer equivalent coverage at lower cost. I switched to the best option, saving you £340 annually. The new policy starts on your renewal date. Here is a comparison of what changed and what stayed the same."
A poorly designed outcome narrative reads: "Insurance policy updated. New provider: Acme Insurance. Annual premium: £1,240." The information is technically accurate. The experience is opaque. The human has no basis for evaluating whether the agent made a good decision, no understanding of the alternatives considered, and no foundation for calibrating their trust in the agent's future decisions.
The Competitive Landscape
The machine customer creates a new competitive landscape in which the traditional advantages of brand, relationship, and distribution are fundamentally altered. A machine customer does not respond to brand advertising. It does not feel reassured by a familiar logo. It does not value a branch network. It evaluates products on their merits - price, terms, coverage, performance - with a thoroughness and objectivity that no human customer can match.
This is simultaneously terrifying and liberating for financial institutions. Terrifying because the competitive moats that have protected incumbent banks for decades - brand recognition, distribution networks, customer inertia - are invisible to machine customers. Liberating because institutions that genuinely offer superior products and terms will be rewarded by machine customers with a speed and decisiveness that human customers rarely demonstrate.
The institutions that will thrive in the age of the machine customer are those that compete on substance rather than perception. This is, arguably, a healthier competitive landscape - one in which the best products win, not the best-marketed products. But it requires a fundamental reorientation of how financial institutions think about their value proposition, their distribution strategy, and their customer experience.
The Design Challenge
The machine customer is not a problem to be solved. It is a condition to be designed for. And the design challenges it presents are among the most consequential in the history of commercial design.
We must design agent identity systems that are robust, verifiable, and interoperable. We must design delegation architectures that allow humans to express their intentions with sufficient precision while retaining sufficient flexibility. We must design outcome narratives that make autonomous action legible and trustworthy. We must design oversight interfaces that give human principals meaningful control without undermining the efficiency that makes machine customers valuable in the first place.
And we must do all of this while the regulatory landscape is still forming - as explored in The Consent Horizon - while the technology is still maturing, and while the competitive dynamics are still being established. The AXD Practice frameworks provide the starting point for this work. The organisations that get this right in the next 18 to 24 months will define the patterns that govern machine commerce for a generation.
The machine customer is coming. The question is not whether your organisation will serve machine customers, but whether it will be designed to serve them well - and whether the humans behind those machines will trust the experience you have built.
The empty hall is filling. The customers have no faces. Design accordingly.