When the Customer Arrives With an Agent

Two possible reactions

Amazon’s decision to restrict Muse is a useful early example. Shopify has chosen integration, PayPal has chosen integration, while Amazon, at least for now, has taken a more restrictive approach.

I suspect many companies will eventually face some version of this decision. They can limit what external agents can see or do in order to preserve control over data, economics, customer relationships or simply the integrity of their platforms. Or they can assume that customers will increasingly arrive through agents and make themselves unusually easy for those agents to understand and operate with.

The real answer will obviously be much more nuanced than “block or open”. Different activities will have different permissions, economics and liability boundaries, but I think the strategic tension is real. It also leads to what I increasingly think of as two different AI waves.

Wave 1 changed who performs the interaction

Wave 1 was ChatGPT, Claude, copilots, chatbots and increasingly enterprise-side agents. For companies, much of the debate became whether AI could answer a request, automate part of a workflow, materially increase employee productivity, handle more contacts differently, and ultimately change operating models, service delivery and pricing.

For service businesses, those are already profound questions. Once productivity becomes visible to clients, AI stops being simply an internal efficiency programme and starts becoming a business-model question.

But one assumption has largely remained intact throughout Wave 1: the customer still reaches the enterprise. There is still an interaction to operate. The person operating it may be human, AI or some combination of both, but the enterprise remains one side of the interaction and the customer remains the other.

Wave 2 may change where parts of the interaction happen

Personal agents introduce something different. Now the AI is not necessarily working for the enterprise; it is working for me.

I can potentially ask it to find me a cheaper insurance policy with equivalent coverage, identify the best fibre subscription available at my address, cancel a subscription I no longer use, understand why my bill increased, or compare several hotels and book the best option based on my own criteria.

The important change is not simply that software performs those tasks faster. It is that meaningful parts of the customer journey can now happen before the enterprise ever sees the customer: research, discovery, comparison, selection and, over time, potentially parts of transaction and servicing.

This is the same basic shift I was trying to describe in The Agentic Enterprise, but Muse makes one part of it much easier to see: the customer-side software layer is beginning to become a consumer product rather than just an architectural possibility.

Wave 1 was mostly about who performs the enterprise interaction — a human or an AI. Wave 2 may increasingly be about where the interaction begins, and which parts of the journey still reach the enterprise directly.

I think that distinction matters enormously for CX.

For CX operators, this could change the perimeter rather than simply automate it

The traditional AI discussion in CX has understandably focused on automation, productivity, human/AI orchestration and new service models. Personal agents add another dimension.

If a customer’s own software begins researching, comparing, preparing requests or completing straightforward actions before reaching the enterprise, the mix of interactions reaching traditional channels may evolve. I do not know yet how large that effect will be, but I think capital markets may eventually ask CX operators a broader question than simply “how much can AI automate?” The more interesting question may become: how does the role of the CX operator evolve when part of the customer journey is increasingly mediated by software acting on the customer’s behalf?

To me, that is a much more interesting question because it does not necessarily imply a smaller role for CX. It may imply a different one.

The second-order effect is where I see the opportunity

Even if personal agents change some traditional interactions, they create an entirely new problem for enterprises.

Imagine a large bank, insurer, telecom operator or retailer with millions of customers. Those customers may eventually be represented not only through familiar digital channels, but through thousands or millions of software agents.

The enterprise then has to answer questions it has barely had to think about before: who does this agent represent, what exactly did the customer ask it to do, what is it authorised to do, which information should it see, which price should it receive, which business rules apply, whether it can execute the transaction directly, what happens when the request is ambiguous, when a human should intervene, who carries responsibility if something goes wrong, and ultimately whether what the enterprise executed actually corresponded to what the customer intended.

This does not look to me like the disappearance of CX. It looks like the emergence of another operating layer.

That may expand what a CX operator can be

Today, very simplistically, a CX operator can be described as an operator of enterprise-customer interactions. The next opportunity might be to operate something broader: the interface between enterprises, customers and the agents acting on their behalf.

That sounds like a relatively small semantic change, but I do not think it is. The capabilities required start touching CX, operations, technology, identity, authorisation, trust, policy, compliance, orchestration, execution and measurement.

The potential customer base is also not one industry. Banks will have to solve it, insurers will have to solve it, telecom operators will have to solve it, and retailers, travel companies, utilities, subscription businesses and probably many others will encounter different versions of the same problem.

Many of them will initially have no obvious operating model for it. This is where I think the opportunity becomes particularly interesting.

The same technology that changes some existing customer interactions may also create demand for a new set of capabilities around agent-mediated customer journeys. For CX operators, that could expand the role from managing interactions to helping enterprises operate an increasingly complex interface between customers, AI agents, enterprise systems and human expertise.

There is also a capital-markets dimension to this

One thing I learned early in Investor Relations is that markets do not wait for companies to fully understand a structural change before beginning to price it. Sometimes the sequence is the opposite.

The market identifies a new theme, starts looking for exposure, and the discussion moves quickly from “is this real?” to “how exposed are you?”, “what changes in your operating model?”, “what capabilities do you already have?”, “where are the new revenue pools?” and “can this become an opportunity rather than simply a new complexity to manage?”

Eventually, if the theme persists, investors want evidence and KPIs.

That is why the recent market reaction is interesting even if some of it ultimately proves excessive. Capital markets are beginning to price the possibility that agent mediation changes enterprise economics before many enterprises even have a system for observing it.

That gap matters.

And this is perhaps the biggest difference between the earlier piece and what I am seeing now. The Agentic Enterprise was primarily about the architecture that might emerge. Muse is an early indication that the economics around that architecture may begin moving before the architecture itself is fully built.

Still early

I may be getting ahead of what is, today, still a very early behavioural shift. Muse is new, consumer behaviour could evolve differently from what early adoption suggests, platforms may restrict agents, regulation may slow some use cases, trust remains a major constraint and business models for personal agents are not settled.

There are many ways the thesis can evolve, but I think the direction is now important enough to watch closely.

Wave 1 was a revolution, and Wave 2 does not start from zero. The models, infrastructure, consumer familiarity and enterprise investment created during the first wave already exist, which may allow the next transition to propagate faster.

That creates some urgency for companies exposed to customer interactions, not in the sense of immediately betting the company on a technology that is still very new, but in the sense of learning before the category becomes obvious.

That means understanding where agents already enter the customer journey, testing what they can and cannot do, seeing what they understand about the company, identifying where current operating models may need to evolve, and working out what should remain closed and what should become agent-accessible.

For CX operators specifically, the more important question may not be whether they can automate more of today’s interaction, but whether they can become part of the infrastructure that helps enterprises operate tomorrow’s.