Who Does Your Agent Work For?

An assistant that shops for you could weaken a business that counts on you not shopping around. It could also create a more powerful business at the entrance to the shop. The distinction became financially consequential this week as Meta's Muse helped divide markets into prospective beneficiaries and casualties of delegated consumer decisions. Reuters reported on September 23 that investors were reassessing companies across commerce and services as the agent gained attention.

The attractive version of this story is that consumers acquire a tireless negotiator. CQ's more demanding thesis is that the real prize will belong to whoever controls the terms on which that negotiator sees, ranks and completes an offer. Cheaper assistance can increase consumer power. It can also relocate the tollbooth. This week's enthusiasm earns serious attention because the capability could be useful; it has not yet earned the conclusion that the resulting bargaining power belongs to the user.

At Connect, Meta announced plans to bring Muse to its AI glasses and introduced the pocket-sized Muse Charm. Its September 24 account of the event also described new ways to interact with the agent. These are product announcements, not an independent demonstration of reliable execution. Their economic significance is the ambition to make delegation an ordinary interface, close enough to use without a separate visit to a website.

A better shopper can change the seller

Imagine a consumer who keeps renewing a service because comparing alternatives takes an evening, a phone call and a tolerance for tedious forms. This is a hypothetical example, but the mechanism requires no futuristic intelligence. If an agent reduces that effort sufficiently, a supplier's ability to retain an inattentive customer becomes less valuable. A rival with a genuinely better offer gains a chance to compete.

This would be a productive use of capability. The consumer gets time back; the better supplier gains a route to demand; the incumbent must improve its offer. A business built partly on customer exhaustion should find the prospect uncomfortable. That discomfort is not evidence that the technology is socially harmful. Sometimes disruption consists of making an existing bargain easier to inspect.

There is, however, a second transaction inside the first. Someone must decide which offers the agent can inspect, how it compares quality with price and what happens when a purchase fails. A cheap option with a punitive cancellation term may be inferior to a dearer one. A recommendation that ignores delivery reliability can save money on the screen and lose it in practice. The user has delegated work, not abolished the need for judgment.

That creates an economic role worth paying for. Maintaining integrations, checking information, handling exceptions and helping reverse an error are real services. A platform that performs them well may deserve a large return. The question is whether its payment rewards those services or rewards its position between a customer and competing sellers. The two sources of profit can coexist, and a polished demonstration does not separate them.

The tollbooth can move closer

An agent that becomes a habitual starting point could make access to the customer more concentrated even while it makes access to information easier. Suppliers might then compete for the agent's attention as well as the user's business. If commercial payments influenced ranking, or if an agent favoured a limited set of integrations, the user could face a narrower effective market behind a more convenient interface.

Those are risks implied by the business structure, not allegations that Muse presently sells undisclosed preference. The distinction is essential. The announcements and market reports do not establish such conduct. They establish enough ambition and investor interest to make the question urgent before a revenue story hardens into an assumption about consumer benefit.

The strongest optimistic answer is competition. A user unhappy with one assistant can choose another. A merchant can distribute through several channels. If agents compete to save users money, an intermediary that quietly worsens recommendations should lose trust and business. Software may make comparison cheap enough to discipline the new intermediary as effectively as it disciplines the old supplier.

That possibility should prevent easy monopoly predictions. Yet it depends on what switching requires. Does the user have to rebuild preferences, permissions and connections? Can a competing assistant examine the same offers? Can the user see why a recommendation won? Convenience that is portable makes competition stronger. Convenience that becomes expensive to leave gives its provider a different kind of advantage.

The point is not that consumers should perform every task manually to remain sovereign. That would defeat the value of delegation. It is that the design of delegation determines whether saved effort becomes freedom to choose or dependence on a default. Access to an agent is only the opening of that bargain.

Capability deserves credit, and a bill

There is also a less glamorous constraint than market power: the cost of delivering competent help. On September 22, Reuters reported an internal Meta test of a human concierge, with contractors handling some calls associated with Muse. The report also described a rollback and Meta's response that any public rollout would require readiness and proper disclosure. This is reported testing, not proof that all agent activity relies on hidden human labour. It is a reason to distinguish an appealing service from a demonstrated model for delivering it at scale.

Human assistance need not discredit a product. Customers may reasonably prefer a service that solves a difficult problem using a person to one that fails autonomously. The economic question is whether the provider can charge enough to cover the assistance and whether the user understands the arrangement. A useful hybrid service and a nearly costless automated service are different investment propositions even when their interfaces look similar.

This is where the week's market excitement needs restraint without cynicism. Useful technology can threaten incumbent margins before its own sustainable margins are known. A falling share price in an exposed industry is not proof of completed displacement. A rising share price in the challenger is not proof that it will retain the value it creates. Investors are pricing possibilities; the underlying businesses still have to earn them.

Oil and bond yields also shaped this week's trading. They affect the cost of funding ambitious products and the budgets of the people meant to use them. They do not answer the narrower question revealed by Muse: when a consumer delegates a decision, which institution gains the power to shape it? That question will survive the next change in the price of crude.

CQ would welcome evidence that agents make users harder to overcharge. Our provisional verdict is that the burden of proof belongs on the claim of transferred bargaining power, not merely on the demonstration of transferred work. Repeated independent comparisons showing lower total costs, successful resolutions and easy switching between agents would shift the judgment decisively toward consumer empowerment. Evidence of paid preference, costly exit or persistent failures concealed by a seamless interface would shift it the other way.

A capable assistant deserves to be paid. A customer deserves to know what it is being paid to optimise. The best new agent will not merely do the shopping. It will make clear whose bargain it is trying to improve.