Over the past year at Silicon Store, we've watched how shoppers ask for things change in front of us. The same user who would have opened five tabs to compare wireless earbuds a year ago now types one sentence to an agent and waits for the answer. Multiply that across hundreds of thousands of sessions and a pattern emerges.
Commerce is moving through three distinct eras — and we're already inside the second one.
The three eras of commerce
Each era names a different relationship between a person and a purchase decision.
In the search economy, the user does the work. They translate a need ("I want a coat that's actually warm, under £150") into keywords ("wool peacoat"), open tabs, compare prices, evaluate reviews, and click checkout. The retailer optimises for clicks; the shopper optimises for whatever time they have.
In the decision economy — where most software has lived for the past year — the model recommends and the user decides. Assistants surface options, summarise reviews, suggest the best fit. The shopper still pulls the trigger, but with fewer tabs and less manual comparison.
In the agent economy — what we've been building toward at Silicon Store — the agent acts on the user's behalf. The shopper states a goal. The agent searches, compares, applies discounts, completes the purchase, and watches the price for seven days afterward. The interface is no longer the storefront. It's the goal itself.
What the shift looks like in production
The most useful thing we can do here is describe what we've actually seen, not what we predict.
We've found that when users speak to an agent instead of a search bar, the requests come out longer and more specific. "Wireless earbuds" becomes "wireless earbuds that fit small ears and don't fall out at the gym." The user no longer has to translate intent into keywords, because the keyword interface is gone.
We've found that the number of vendor pages a user opens themselves drops sharply within a few sessions of trusting an agent — not because the user becomes lazy, but because the agent surfaces what they need without the manual click-through.
We've found that the metrics that mattered in the search economy — click-through rate, session depth, time on site — become almost unrelated to whether a purchase actually happens. A successful agent interaction is short, narrow, and ends with a decision. That's exactly the inverse of a successful search interaction.
And we've found that when something goes wrong — a return, a price drop, a wrong size — the agent handles it inside the same conversation that started the order. The customer doesn't "go back to the website." In a lot of cases, there is no website to go back to.
When AI becomes the customer
The downstream effect of this shift is that merchants are increasingly selling not to people but to the agents acting for them.
Concrete things that start to matter more:
- Structured product data instead of marketing copy. An agent can parse a spec sheet; it cannot be persuaded by a hero banner.
- Verifiable quality signals instead of one-line testimonials. Reviews with weights and dates beat reviews with stars and emojis.
- Machine-readable policies instead of fine print. A return policy an agent can quote back to the user before purchase is a competitive advantage.
- Real-time pricing endpoints instead of cached pages. Agents compare actual checkout prices, not list prices.
We see this pattern playing out on our own platform. When our agent ranks options, sellers with complete structured data are consistently surfaced ahead of sellers with prettier websites but ambiguous spec sheets — even when the user has no preference between them. The agent picks the one it has more information about, because information is what reduces decision risk.
This is not unlike what happened to SEO twenty years ago: a structural shift in how discovery worked rewired what counted as "good marketing." The agent economy is doing the same thing again — but for purchase intent rather than top-of-funnel attention.
A tradeoff worth naming
Every era has a tradeoff that defines it. In the search economy, the user trades time for control. In the decision economy, the user trades some control for less time. In the agent economy, the user trades the remaining control for an outcome — and that's a real loss, not just a UX upgrade.
There is a tradeoff: more autonomy buys faster, cheaper, better-fit purchases. But it also means the user gives up some of the small decisions that made shopping feel like shopping. Picking the specific shade of blue. Spotting the model with the colour you didn't know you wanted. The serendipity of the wrong tab on a Tuesday night.
We don't think this is necessarily worse, but it's different — and it's worth being honest that the agent economy isn't a strict upgrade of the search economy. It's a different deal, optimised for different things.
Human agency moves up a level
A common mistake about agent economies is to assume that handing off execution means handing off agency. That isn't what happens in practice.
What we've found is that agency moves up a level. Users stop deciding "which earbuds" and start deciding "how strict on price," "how much to prioritise return window," or "what's the maximum I want to spend on this category this month." The decisions get higher-leverage but less frequent.
This is the same shift that happened when programming moved from punched cards to compilers to high-level languages: less of the human's attention goes to the mechanics, more goes to the goal. The work doesn't disappear; it relocates.
Caveats
There are real things this shift doesn't change.
It doesn't change the underlying economics of physical goods. An agent can find a better price, but it cannot manufacture a coat for less. The savings the agent surfaces are real, but they come from compression of the price-discovery layer, not from creating value at the source.
It doesn't eliminate the need for human judgement on novel purchases. The first time you buy something you've never owned before, the agent's recommendation is a starting point — not a substitute for trying it. We've found that agent-mediated purchases convert most cleanly in categories where the user already has prior experience and clear criteria.
And it doesn't make every category equally agentic. Categories with high specification clarity — electronics, household replenishment, books — move into the agent flow quickly. Categories that are about taste, fit, or feel — fashion, beauty, art — stay closer to the decision economy: the agent surfaces options, but the human still picks. We don't expect this gap to close on a uniform schedule, and we're sceptical of anyone who tells you otherwise.
Looking ahead
The companies that win the agent economy won't be the ones with the smartest agents. They'll be the ones that build infrastructure for a world where most purchases stop going through a screen.
For shoppers, that means agents that explain themselves, never overstep, and can be wound back to manual at any point. For merchants, it means catalogues an agent can parse without a browser. For payment networks, identity systems, and trust infrastructure, it means handling transactions where the buyer is software acting under explicit human authority.
This is the future we're building toward at Silicon Store, and we'll keep writing about what we're learning as we go. The agent economy is still early; we'd rather be honest about what's working and what isn't than promise the next era is arriving on a fixed timeline. If you're building in this space, we'd love to hear what you're seeing — what surprised us most over the past year was how much the right pattern looks different in production from how it looks in a launch post.