The Agentic Customer Has Entered the Chat

For years, companies have used automation to manage customers. To route them. Deflect them. Predict them. Nudge them. Contain them. Sell to them more efficiently. Sometimes, if we are being very generous, even help them.

Now the customer is getting an agent of their own.

Not only a chatbot that answers a question, but an AI that can research, compare, filter, recommend, negotiate, transact, switch providers, and eventually manage more of the admin of ordinary life on the customer’s behalf. The language around this is still settling, but the direction is clear enough: the agentic customer has entered the chat. Accenture now describes agentic commerce as a “generational reset” in who and what selects and buys products, while McKinsey notes that consumers are already moving along an automation curve in which AI increasingly mediates shopping decisions, even when the human customer remains the one with the final intent. (Accenture)

That idea has been scratching at my thoughts for days.

Because this is not only a retail story. It is not only about whether an AI assistant will choose your toothpaste, compare your insurance, or remember that you always buy the same dog food but object on principle to paying full price for it.

It is a customer power story.

For decades, organisations have benefited from customer inertia. People stay with the bank they opened at university, the insurer they once compared in a hurry, the telecom provider whose tariff they no longer understand, the subscription they meant to cancel three years ago, or the brand they vaguely prefer because switching is tedious and life is already full of laundry.

Agentic AI threatens that comfortable fog.

If your customer’s assistant can read the fine print, compare offers, check reviews, weigh trade-offs, and move them to a better option before breakfast, then loyalty built on friction is about to look very different. In Australia, recent survey data found that 73% of consumers had already used AI for product research and nearly half trusted AI agents to transact on their behalf. In North America, more than half of frequent AI users say they have tried a product based on an AI recommendation, and 90% say they would be open to switching from a preferred brand if their AI assistant found a better alternative. (The Australian)

The future is not less human. It is more deliberately human.

And sometimes, being deliberately human means noticing when a customer is about to arrive with a much better memory, a much lower tolerance for nonsense, and a small digital bloodhound at their heel.

The Customer Is Still Human. The First Judge May Not Be

It is tempting to frame this as a story about customers disappearing behind machines. I do not think that is quite right.

The customer is still human. They still have needs, values, fears, preferences, budgets, habits, quirks, loyalties, and moments of weakness involving cheese they did not come in for. What changes is that the first layer of interpretation may increasingly be handled by an agent acting on their behalf.

Before the human customer sees your offer, their AI may already have compared your price, return policy, delivery promise, contract terms, complaint history, sustainability claims, cancellation process, and whether your competitors offer materially better value. McKinsey’s research in Europe suggests that AI is already influencing the decision stage, even where full autonomous purchasing has not yet become mainstream; before a consumer buys, an AI tool may have narrowed the field and explained why certain options did not make the cut. (McKinsey & Company)

That is a very different front door.

Businesses have become skilled at designing for human attention. Search ranking. Brand recall. Emotional cues. Loyalty programmes. Beautiful storytelling. Friction placed so gently in the cancellation journey that nobody calls it friction in the meeting.

But an AI agent does not care that your campaign had a delightful soundtrack. It does not feel nostalgic because your packaging looked like childhood. It is unlikely to be seduced by three paragraphs of lifestyle language before discovering the actual price on the fourth scroll.

It will ask different questions.

Is the offer legible? Are the terms clear? Can the value be compared? Is the policy fair? Can the transaction be completed cleanly? What happens if something goes wrong? Is this brand a good choice for the human I represent?

A person may forgive a little ambiguity if they like you enough. A machine is far less likely to be charmed by your adjectives.

We May Be Entering the End of Profitable Confusion

There is a deliciously uncomfortable possibility here. A great many business models have been quietly subsidised by the fact that customers do not have infinite time, perfect recall, or the will to inspect every clause before clicking “agree.” Some confusion is accidental, born of old systems, tangled ownership, and policies that accumulated like sediment. Some is more strategic than that.

Either way, agentic customers may make profitable confusion much harder to maintain.

Banks are already watching this closely. McKinsey notes that 23% of consumers in its Global Banking Annual Review 2025 survey were using generative AI for financial tasks at least monthly, including product understanding, investment advice, and comparisons. The concern for banks is not only that AI may help customers make choices. It is that customer relationships may be disintermediated when an external agent becomes the trusted layer through which those choices are made. (McKinsey & Company)

The same logic travels easily into insurance, utilities, telecoms, travel, retail, subscriptions, healthcare, and any industry where customers have historically had to tolerate complexity because the cost of understanding it was too high.

A customer-side agent can keep track of renewal dates, compare usage patterns, identify wasted spend, surface better options, and recommend action before the human customer has even remembered that their contract exists. It can notice that the “loyalty discount” is worse than the new customer deal. It can count the number of times a company failed to deliver against its promise. It can decide that the warm brand feeling is no longer enough to offset the cold mathematics.

That does not mean every customer will become purely rational. Humans have never been only spreadsheets with hair. But it does mean brands may lose some of the protection previously afforded by forgetfulness, inconvenience, and overwhelm.

The companies most exposed will be the ones whose apparent loyalty was really exhaustion in a cardigan.

Your Brand May Soon Have Two Audiences

This is one of the most useful shifts in the current conversation.

Accenture argues that brands now have two audiences in commerce: the person and the agent acting on that person’s behalf. Deloitte similarly frames agentic commerce as a new channel in which AI agents collaborate directly with retailer systems to fulfil demand, while McKinsey points out that retail data will need to become increasingly machine-readable as shopping moves through higher levels of delegation. (Accenture)

That does not mean we stop writing for people and start publishing love letters to APIs.

It means we need to become much clearer about the difference between being appealing and being understandable.

The human customer still needs meaning, reassurance, desire, delight, story, trust, and a reason to choose you when the answer is not purely numerical. But their agent needs clean product data, unambiguous policies, structured information, transparent pricing, accurate availability, interoperable systems, and fewer places where the truth is hiding behind a dropdown menu.

For a while, businesses have been able to confuse digital experience with digital theatre. Beautiful interface. Warm copy. Clever nudges. Meanwhile, the actual terms are buried, the handoff is clumsy, and the customer has to leave the journey to find out what should have been obvious at the start.

Agentic customers will be less tolerant of that split. A human might say, “I like this brand, let me see.” Their AI may say, “This brand is 18% more expensive, has worse cancellation terms, and has generated repeated complaints about returns. Shall I show you better options?”

That is not the death of brand. It is the death of brand as camouflage.

The Customer Experience Will Need to Survive Translation

This is where my own customer service brain gets very interested. What happens when your customer journey is no longer experienced only by a person, but also interpreted, summarised, and represented back to them by their own AI?

If your returns process is fair but hard to understand, the agent may mark it as cumbersome. If your policy is technically compliant but emotionally cold, the human may only encounter the condensed verdict: “This provider offers limited flexibility.” If your service team repeatedly fixes the same upstream problem with grace, but the system itself remains broken, the agent may not care that your people are lovely. It may simply learn that this brand creates avoidable friction.

That last point matters. Customer service has long been where organisations rely on humans to rescue the experience from the design. A thoughtful agent, a good escalation specialist, a manager with enough judgement, a frontline professional who knows how to turn a poor process into a tolerable outcome. That human repair work has protected many brands from the full consequence of their own complexity. But an agentic customer may assess the system before the human rescue ever has a chance to shine.

This is not an argument against warmth. It is an argument against using warmth as a debt collector for bad design.

If your experience only works because skilled people repeatedly compensate for what the system gets wrong, the customer’s AI may eventually conclude that your business is inefficient, unpredictable, or simply less attractive than the alternative.

The future customer journey will need to be not only emotionally intelligent for people, but structurally intelligible to machines. That is a different design brief.

The New Loyalty Test

We talk a great deal about customer loyalty, but much of what we call loyalty has been an unstable blend of habit, inertia, effort avoidance, and partial satisfaction. Agentic AI may help us discover what loyalty is worth when switching becomes easy. That is not necessarily bad news for good companies. It may be rather good news indeed.

If your product is genuinely strong, your value is clear, your policies are fair, your service is dependable, and your customers trust you for reasons deeper than friction, then an agent that compares options may keep recommending you. Accenture’s research suggests that brands optimised for both humans and agents could become more visible and more competitive, not less. (Accenture)

The companies in trouble are those that have mistaken customer difficulty for customer devotion.

If someone stays because they love what you do, that is loyalty. If someone stays because cancelling requires three phone calls, a password reset, and a level of emotional fortitude usually reserved for visa applications, that is captivity with a monthly fee.

Agentic customers will make that distinction harder to ignore.

The Risk: A World of Perfectly Optimised Bad Choices

We should not become starry-eyed about this either.

Agents acting on behalf of customers will raise serious questions about trust, bias, privacy, accountability, commercial influence, and the invisible objectives embedded in the systems doing the recommending. Deloitte warns that AI agents are scaling faster than many enterprises’ guardrails, with only 21% of surveyed organisations reporting mature governance for them. Regulators are already concerned in sectors such as banking, where consumer-facing agentic AI raises questions about responsibility when autonomous systems make or influence decisions with real financial consequences. (Deloitte Brazil)

A customer’s AI is not automatically a wise little owl perched on their shoulder.

Who trained it? What incentives shape its recommendations? What data does it have? What does it optimise for? Price? Convenience? Profit to the platform? Stated preference? Inferred behaviour? Does it understand that the cheapest option may be disastrous for a vulnerable customer? Can it explain why it chose what it chose? Who is accountable when the answer is confidently wrong?

We have already seen that current AI shopping assistants are not always the elegant digital butlers imagined in the glossy reports. Recent reporting has highlighted retailer bots that become intrusive, scripted, or simply bad at basic customer interactions when the desire to sound human outruns the discipline of being useful. (The Guardian)

So no, this is not a story about machines becoming perfect customer advocates while businesses quake in their loafers. It is a story about the customer side of the equation becoming more technologically capable too. And that changes the balance of power.

What Deliberately Human Organisations Should Do Now

The easy response would be to create a workstream called “agentic commerce readiness,” appoint twelve people to it, and produce a deck featuring a lot of arrows.

The more useful response is to ask harder questions about your existing business. Can your offers be understood without persuasion doing all the work? Are your policies clear enough to survive summarisation? Would a neutral comparison engine recommend you for the reasons you hope customers choose you? Where are you relying on habit, opacity, or customer exhaustion instead of value? What parts of your customer journey only function because humans downstream keep rescuing them? If an AI agent reviewed your complaints, service failures, refund friction, and cancellation process, what would it conclude about your organisation?

Those are not only technology questions. They are trust questions. The businesses that will do well in this next phase are not merely the ones with the cleverest customer-facing agents of their own. They will be the ones who become easier to understand, easier to compare fairly, easier to transact with, and harder to dismiss on substance.

They will make machine-readable clarity and human-readable dignity part of the same design problem. They will stop assuming that every customer journey begins with a human looking at a homepage and ends with that same human filling in a checkout form. They will recognise that a customer may soon arrive through a proxy that has already formed a view. And perhaps most importantly, they will stop asking only, “How do we use AI to know the customer better?”

They will also ask: “What happens when the customer uses AI to know us better?”

The Customer Question Is for Everyone

The question for this article is not only for marketers, retail leaders, or digital teams. It belongs to anyone who designs a customer journey, owns a policy, writes product information, builds a chatbot, sets a price, measures loyalty, manages service recovery, or assumes that customers will keep accepting friction because they always have.

How would your business look if the customer had perfect recall?

What if they could compare every promise you make with every experience you actually deliver?

What if they no longer needed to remember when to review, renew, switch, escalate, cancel, or ask whether there is a better option elsewhere?

Would your value become clearer?

Or would your weaknesses become impossible to hide?

The agentic customer is interesting not because it removes the human from the story, but because it may reveal how much of modern customer experience has depended on the human being tired, distracted, trusting, busy, or simply not having the time to inspect the system properly.

That is an uncomfortable thought. It is also a useful one.

Conclusion: The Customer Has a Co-Pilot Now

The dominant customer experience story has been about what companies can know, predict, automate, and optimise about the people they serve. The next chapter may be about what people, with the help of their own agents, can know, predict, and optimise about the companies asking for their money. That shift will not destroy humanity. It may, in fact, force a cleaner version of it.

Less seduction by confusion. Less loyalty built on inconvenience. Less warmth used to cover for poor design. More clarity. More fairness. More genuine value. More trust that can survive inspection.

The future is not less human. It is more deliberately human. And perhaps the most human thing a company can do in the age of agentic customers is stop hoping the customer will not notice, and start becoming genuinely worth choosing when they do.

If your customer arrived tomorrow with an AI assistant that could compare your promises, prices, policies, complaints, and competitors in seconds, what would it tell them?

Where are you relying on customer effort, opacity, or inertia instead of real value?

And what would you need to change so that both the person and the agent acting on their behalf could recognise your organisation as the right choice?