The End of the Click

For most of the last twenty years, digital commerce has been framed as progress. More choice, more speed, more access, more convenience. And in many ways that is true. But there is a quieter reality underneath it, one that most people feel even if they do not articulate it clearly. The cost of modern commerce is not primarily financial. It is cognitive. The work of deciding has been pushed onto the individual, and the systems we built have normalised that as the price of participation.

Search was meant to be the solution. You type what you want, the internet returns the world, you pick the best option. But search has become a proxy battle between platforms, advertisers, and optimisation tactics. The “results” are no longer a neutral mirror of what is available, they are a negotiation between incentives. In parallel, social media became the discovery layer, not because it was inherently better at matching intent, but because it was better at shaping desire. Somewhere in the middle of those two forces, people ended up doing an enormous amount of work just to complete relatively simple tasks.

You can see it in everyday behaviour. Someone discovers a product on Instagram, then opens Google to check if it is real, then searches Reddit to see what people actually think, then checks prices on Amazon, then goes back to the brand site, then gets interrupted, then forgets where they were, then does it again later. This is not a user journey, it is a patchwork of systems that do not talk to each other, and it turns the consumer into the glue.

This is why “choice” has quietly become a burden. The infinite shelf sounds empowering, but in practice it produces decision fatigue. Not because people dislike having options, but because most options are poorly described, poorly differentiated, and wrapped in marketing language that makes verification difficult. In B2B, this shows up as error rates and friction. In consumer behaviour, it shows up as abandonment. People do not always abandon because they do not want the product. They abandon because the process of getting to a confident decision is too expensive in time and attention.

At the same time, trust has fragmented. Consumers interact with social platforms constantly, but they do not trust them. They distrust brand claims, they distrust paid placement, they distrust reviews that feel purchased, they distrust marketplaces that feel rigged, and they distrust the sense that “the best option” is simply the one with the biggest budget. So the act of shopping becomes less about finding something and more about triangulating truth.

This is the context in which agents appear, and it is why I think the shift is deeper than “AI in shopping.” It is not simply a layer of convenience on top of the existing model. It is a different behavioural posture. It is the move from discovery to delegation.

Delegation is not just outsourcing labour, it is outsourcing the decision-making process itself, in whole or in part. Instead of scrolling through options, you instruct. Instead of comparing ten tabs, you define constraints. Instead of hunting for proof, you expect the system to bring you validated signal. The user is no longer navigating an interface. They are expressing intent, and a system is acting on their behalf.

This is also why the “interface” conversation is often a distraction. People talk about chat, voice, smart glasses, and new devices, and all of that matters, but it is not the centre of gravity. The real shift is architectural. When commerce becomes agent-mediated, the question changes from “how do I browse” to “who do I trust to act.” That is a fundamentally different market dynamic.

It also plays out differently across cohorts and regions, which matters because agentic commerce will not be adopted uniformly.

Gen Alpha will likely treat delegation as normal, because they are growing up in a world where voice and visual input are native, and where the expectation is that systems should infer intent without friction. They will not understand why they need to learn the structure of websites or filter menus. That model will feel as archaic to them as dial-up feels to us.

Gen Z is more complex. They are price sensitive and value obsessed, but not in a simple “cheap versus premium” way. They trade down and they splurge, often in the same basket, because what they are optimising for is emotional value and identity, not just function. They also do not search in the traditional sense. They search socially. They want “vibes,” proof, context, community validation. So an agent that behaves like a sterile procurement tool will not land. An agent for this cohort needs to understand social proof, cultural relevance, and the way desire is formed, not just the specs.

Millennials, meanwhile, are often operating under time scarcity. Shopping is not entertainment, it is logistics. They are already comfortable with subscriptions, BOPIS, automation, and managed flows. For them, an agent becomes a chief of staff, a system that reduces decision overhead and removes repetitive work. If it proves reliable, they will happily hand over low-risk purchases and routine tasks.

Gen X and Boomers, who still control an enormous share of spending power, tend to be less interested in novelty and more interested in clarity and accountability. They do not necessarily reject automation, but they reject opacity. They want to know why something was recommended, what assumptions were used, and how to escalate to a human when it matters. For these groups, an agent must behave less like a black box and more like a transparent concierge.

These nuances matter because they point to what the “agentic interface” actually needs to be. It cannot be one thing. It will be shaped by trust thresholds, cultural norms, and the reality that not all purchases carry the same risk.

Which brings me to what I see as the core design principle for the next era. The future is not a binary leap to full autonomy. It is progressive autonomy. People will not suddenly allow systems to spend freely on their behalf. They will, however, increasingly allow agents to do the work, and then ask for a single approval moment. Over time, as reliability is proven in low-risk contexts, autonomy expands.

You can think of it as three modes.

In high-risk purchases, the agent is an advisor. It researches, compares, and cites. It reduces noise and makes reasoning visible, but the human decides.

In medium-risk purchases, the agent is an intern. It drafts the plan, fills the cart, assembles the options. The human reviews and approves.

In low-risk, repeat purchases, the agent becomes a proxy. It executes, and the “interface” is essentially a notification, an undo, or a summary.

This is where many current discussions about shopping assistants miss the real story. If the system cannot explain itself, it will not earn trust. And if it cannot recover from mistakes, it will not be tolerated. In an agentic world, usability is still important, but it is subordinate to what I would call trustability. Explainability, steerability, and reversibility are not extras. They are the product.

This shift also forces brands and merchants to face an uncomfortable truth. The battle is no longer for the user’s click. It is for the agent’s confidence.

That changes how brands “show up” online. Marketing language is not useful to an agent. Poor product data is fatal. If your catalogue is not structured, if your attributes are vague, if your availability and terms are unclear, the agent cannot reliably select you. SEO becomes less important than what I would call machine legibility. Agents need clean, consistent, high-fidelity information that can be verified, compared, and acted upon.

And it also changes the role of storefronts. In an agentic environment, the transaction often does not happen on the brand website in the traditional sense. Commerce moves to the edge, into the chat window, into voice, into messaging threads, into interfaces that are generated on demand, used once, and discarded. Your “store” becomes less of a destination and more of a set of capabilities that can be invoked anywhere.

If this sounds abstract, the implications are actually simple. Most commerce volume will become quieter and more automated. The boring categories, replenishment, basics, commodities, will drift towards invisible buying, with humans intervening only when something changes or goes wrong.

But the interesting categories, fashion, design, gifting, identity-driven purchases, do not disappear. They become more immersive, more collaborative, more augmented. Agents will not replace taste, but they will accelerate exploration and reduce friction, acting as creative partners rather than search tools.

So the shift is not “shopping goes away.” It bifurcates. One part becomes chore. The other becomes experience. Both become agent-mediated, but in different ways.

And that is the real horizon I take. We are not simply building smarter search. We are building the infrastructure for delegation. The networks that win will not be those that amplify noise or maximise browsing time. They will be those that respect attention, reduce cognitive load, make reasoning visible, and let individuals express intent clearly while allowing brands to respond with precision.

The internet is not becoming quieter. But people are becoming more selective, and they are increasingly willing to delegate the noise away.

That is what is changing. And that is where the architecture of autonomous commerce actually begins.