Chapter 2: The Recommendation Economy

Ask most people whether they delegate judgment to machines and they will say no. They use tools. They rely on data. They make their own decisions. But then ask them how they found their last hotel, which route they took to an unfamiliar address, how they chose a contractor after a quick search, what prompted them to watch the show they finished last week. The answers tend to reveal a relationship with algorithmic recommendation that is considerably more embedded than the first answer suggested.

We are already living inside a recommendation economy, and have been for long enough that the experience feels native rather than novel. The shift from looking things up to being shown things happened gradually, and the version most people are aware of, the helpful suggestion, the relevant result, the curated feed, is only the surface layer. Beneath it, the same systems have been quietly extending their reach into decisions that carry considerably more weight than which television program to watch.

When Advice Becomes Judgment

There is a meaningful difference between information, recommendation, and judgment, even though the three blur into each other in practice. Information is raw material: a list of flights between two cities, the menu of a restaurant, the candidates for a position. Recommendation is a filtered version of information: these three flights are most likely to suit you, this restaurant is most likely to match your preferences, these five candidates are worth reviewing first. Judgment is what comes next: the actual selection, the weighed decision, the choice that commits resources, time, or trust.

The line between recommendation and judgment has been moving for years. When a GPS navigation system shows you one route rather than presenting all available options with equal prominence, it is making a judgment call about what matters, not just presenting information. When a search engine surfaces ten results from among billions, the ranking is a judgment about relevance and authority that shapes what most people will ever see on a given topic. When a hiring platform screens applications before a recruiter reviews them, the judgment about who is worth a human's attention has already been made by the time a human gets involved.

None of these are presented as judgment. They are presented as tools, features, assistance. The framing matters because it shapes how critically people engage with what they receive. A recommendation you understand to be a recommendation you can evaluate and push back on. A result that feels like the answer to a search, or a list that feels like the obvious starting point for a decision, is harder to interrogate because it does not announce itself as a judgment call at all.

The Domains Where Delegation Is Already Mature

The recommendation economy is not concentrated in entertainment and retail, though those were its early proving grounds. Over the past decade it has extended into domains where the stakes of any individual decision are substantially higher and the consequences of poor recommendations are considerably harder to reverse.

Travel and logistics were among the first domains to see widespread delegation. Navigation systems replaced not just maps but the spatial reasoning and local knowledge that map-reading develops. People who grew up routing their own drives have a residual orientation toward place and direction that younger people who have never navigated without GPS often lack. The delegation was efficient. It was also not free.

Healthcare triage and clinical decision support have incorporated algorithmic recommendation at multiple points in care pathways, from the symptom checker that suggests whether to seek care, to the risk stratification tools that determine how urgently a patient is seen, to the diagnostic decision support that flags possible conditions for physician review. Each of these is framed as assistance to clinical judgment rather than a replacement for it. The distinction is genuine and important. But it is also one that erodes under pressure: when a clinician is seeing forty patients in a shift, the algorithmic recommendation that would take ten seconds to override may simply get accepted instead.

Financial services have incorporated recommendation systems into lending decisions, investment allocation, insurance pricing, and fraud detection. The credit scoring model that determines whether a mortgage application proceeds represents a judgment that was once made by a loan officer who could be asked questions, who could weigh context, and who could be held accountable for a decision. The model produces a number. The number produces a decision. The person whose application was declined has very limited insight into what drove the outcome and limited recourse to challenge it.

Hiring and employment represent perhaps the most consequential domain of current recommendation economy maturity. Applicant tracking systems, resume screening algorithms, and candidate scoring tools are now standard infrastructure in most large organizations. The practical effect is that a significant portion of hiring decisions are made by systems that a candidate never interacts with and whose criteria they cannot inspect, before any human being has read their name. The downstream effects on whose careers advance, whose applications disappear, and which kinds of backgrounds get consistently filtered out are real and have been documented, even as the systems generating those effects remain largely opaque to the people they affect.

The Continuum Most People Have Not Mapped

One of the more useful ways to think about where any given recommendation sits is along a continuum from suggestion to delegation to automation. At the suggestion end, the system offers an option and the person actively chooses whether to take it. The friction of choice is preserved. At the delegation end, the system makes the decision and the person ratifies it, usually without close examination. At the automation end, the system makes the decision and acts on it without the person being involved at any stage.

Most people, if asked, would place themselves primarily in the suggestion and recommendation stages. The more accurate picture for many decisions, particularly habitual ones, is that they have slid into delegation without a conscious choice to do so. The default option gets accepted, the top search result gets clicked, the subscription auto-renews, the routing app's first suggestion gets followed. The slide from recommendation to delegation does not require a decision to delegate. It only requires the absence of a decision to do otherwise.

What Convenience Has Cost

The efficiency gains from the recommendation economy are real and should not be dismissed. Routing around traffic saves time that matters. Finding relevant medical information quickly has genuine value. Surfaces that surface the most likely match for a given need reduce the cognitive overhead of daily decision-making in ways that free up attention for things that actually warrant it.

But alongside those gains, something has been quietly depleted. The habits of mind that developed around making decisions without algorithmic assistance, the judgment about which route to take, which source to trust, which candidate to consider, which vendor to engage, have atrophied in the same way that physical capacities atrophy when machines take over the work those capacities were built for. The atrophy is invisible while the system is working well. It becomes visible when the system is wrong, unavailable, or has been manipulated, and the person who trusted it discovers they no longer have the independent judgment to detect the problem.

That is the setup for what Chapter 3 examines: what happens when the recommendation layer does not just facilitate decisions but becomes the infrastructure that everything else depends on, and what it means that this infrastructure can be engineered, gamed, and exploited by anyone who understands how it works.