Superagency

The book's strongest and most reusable idea is that the core design question for AI is not how smart will it get, but what does it let people do more effectively, more confidently, or more independently. That sounds almost simple, yet it is the move that saves the book from becoming another generic future-of-work tract. Hoffman and Beato are trying to redirect attention away from spectacle. They are less interested in machine intelligence as a marvel than in agency as an outcome. That means they care about whether people can make better decisions, access more capability, navigate more complexity, and act with more leverage in daily life.

The authors make this credible by repeatedly contrasting agency with familiar fears. If AI is framed only as labor replacement, then the discussion narrows to jobs, wages, and displacement. If it is framed only as existential risk, then the discussion becomes abstract and often frozen. Agency gives the book a way out of that stalemate. It shifts the focus to what users, citizens, workers, and institutions can actually do with better intelligence layered into their lives. The authors push this through examples that range from tutors and legal advice to mental health support, scientific discovery, and entrepreneurship.

What matters for practitioners is not to take this as an article of faith. Agency can become a noble-sounding label slapped onto almost any AI proposal. A chatbot that adds noise to a workflow can be described as empowering. A recommendation engine that narrows judgment can be described as helping. That is why the book's framing is useful only when paired with sharper planning discipline. If a team says it wants to expand user agency, it should then be forced to answer: for whom, in what moment, against what current constraint, and in what observable way? The book itself does not always do that work. But the concept remains powerful because it gives teams a better north star than efficiency alone. It pushes discovery toward decisions, guidance, and leverage, which is a better starting point than feature excitement. In Waypoint terms, this is less a proof than a framing lens. Used well, it helps teams define better future states. Used badly, it becomes rhetoric.

The chapter on Big Knowledge contains perhaps the book's clearest bridge from macro argument to practical value. Hoffman and Beato argue that modern life is not defined simply by scarcity of information but by absurd abundance. The amount of data, media, research, commentary, signals, and telemetry in circulation now exceeds any individual's ability to absorb or use it meaningfully. Their point is not merely that there is a lot to read. It is that the human bottleneck has moved. For many consequential decisions, the problem is not that relevant information does not exist. The problem is that it is too dispersed, too voluminous, too contradictory, and too fast-moving for ordinary human cognition to synthesize.

That is the context in which AI becomes meaningful. Its promise is to turn Big Data into Big Knowledge, not by solving truth perfectly, but by making the world more legible and more navigable. The authors build this by revisiting older fears about centralized information systems. Orwell looms over the chapter, as does the failed proposal for a national data center in the United States in the 1960s. These examples let the book acknowledge a persistent anxiety: more information power can easily look like more surveillance and control. Yet the authors want to show that the historical arc did not end in pure domination. Computing and networks also helped generate more individual freedom, more creativity, more pluralism, and broader access to knowledge.

This is where the book starts sounding genuinely useful for discovery practitioners. If AI's biggest practical contribution is sense-making under conditions of overload, then the best opportunities are not necessarily in automating whole roles. They are in reducing uncertainty, surfacing relevant context, synthesizing options, and helping people make better calls in moments that matter. The likely misapplication is to treat the existence of too much information as automatic justification for AI intervention. Many environments contain lots of data but poor governance, poor process, weak incentives, or ambiguous goals. In those cases, an AI layer may magnify mess rather than create knowledge. Still, as a planning frame, this idea is strong: look for domains where people are under-informed not because nothing is known, but because too much is known badly.

One of the book's more provocative moves is its resistance to what it calls problemism, the tendency to see technology mainly as a source of likely harm. This is not the same as denying that harms exist. Rather, the authors argue that modern technology debates often become so saturated with downside scenarios that they fail to articulate what a genuinely better future would require. In that vacuum, the status quo gains an unearned moral advantage. Existing failures start to look safer simply because they are familiar.

The examples used to carry this argument are revealing. Mental health support is the most striking. The authors know this is a fraught domain. AI in therapy-like contexts sounds reckless to many readers. But they use that discomfort to make a more forceful point: the status quo is not neutral. Many people have no access to therapy, delayed access, unaffordable access, or inconsistent human support. Under those conditions, dismissing AI support on principle can mean defending scarcity rather than safety. The same logic appears in discussions of tutoring, scientific research, entrepreneurship, and other domains where capability is unevenly distributed.

This is one of the book's most important correctives, especially for planners. Many AI conversations are dominated by the imagined harms of deployment while saying very little about the harms of non-improvement. A service that currently leaves users confused, unsupported, or badly informed is already producing damage. If AI could reduce that damage, even imperfectly, then the comparison set changes. That is a valuable planner insight. But it is also where the book risks overreach. Just because the status quo is flawed does not mean an AI intervention is good, ready, or accountable. The useful part of the idea is its insistence that better futures need to be described in enough detail to be designed toward. Before critique dominates the room, ask what meaningful improvement would look like, for whom, and compared to what current failure. Once that is named, then assumptions, risks, and trade-offs can be examined more honestly.

A major through-line of Superagency is its defense of iterative deployment. Hoffman and Beato use the term in a broader sense than typical product teams might. They do not mean simply releasing a beta and patching bugs. They mean that AI development, public adaptation, trust formation, and even regulatory understanding should occur through phased exposure to real-world use rather than through closed-door expert control alone. In the authors' telling, OpenAI's public release of ChatGPT is the emblematic case. Instead of waiting for a perfectly safe, fully understood endpoint, a bounded version of the technology is put into public hands. Users encounter it, react to it, misuse it, benefit from it, complain about it, and thereby generate both product learning and social learning. Trust, they argue, emerges through consistency over time and through widespread familiarity rather than through theoretical assurances.

This idea becomes more developed when the book moves into benchmarking and governance. The authors are good at noticing that much public AI debate uses the language of arms races and apocalypse while giving less attention to transparent performance comparison. Their discussion of benchmarking, model evaluation, and something like Regulation 2.0 is one of the more concrete parts of the book. The underlying claim is that public transparency about what models do well, where they fail, how often they hallucinate, and which tasks they should or should not be trusted for can create a more participatory and evidence-rich governance environment.

For Waypoint practitioners, the useful takeaway is that legitimacy and safety do not arrive only through policy documents. They are also produced through bounded use, visible performance, and explicit learning loops. In planning terms, that means AI work should be framed as an exploration with designed feedback, not as a one-time declaration of readiness. The limitation is obvious. Iterative deployment is easier to defend when you are close to frontier development and relatively trusting of commercial actors. It is less obviously sufficient in high-stakes domains, power-asymmetric settings, or contexts where users are not meaningfully positioned to consent or contest. But as a planner idea, this section is better than most AI boosterism because it at least recognizes that adaptation, testing, and public legibility are central, not optional.

The book's most memorable metaphor is informational GPS. Hoffman and Beato use GPS as a precedent for what broadly distributed AI might become: a form of invisible but powerful guidance infrastructure that helps people navigate complexity in real time. GPS matters in the book because it was once expensive, limited, and military, and later became a widely available public utility layered into everyday life. It does not decide every destination for you, but it radically lowers the cost of orientation, rerouting, and wayfinding. That is the authors' model for AI at its best. Not an all-controlling intelligence but a guidance layer that helps people find better routes through knowledge, decisions, and services.

What makes this metaphor more than a slogan is the way the authors connect it to networked autonomy. They argue that much of modern digital value comes from resources that are privately built yet broadly usable: maps, reviews, payment data, platforms, public information, routing tools, and countless other infrastructures that can be recombined into more personalized support. The networked autonomy argument deepens the claim by refusing a simplistic freedom story. The chapters on cars, roads, laws, and the Interstate Highway System argue that freedom often increases not when constraint disappears, but when infrastructure and rules make higher-order agency possible. Driving is empowering partly because roads exist, standards exist, signals exist, safety expectations exist. The provocative implication for AI is that greater individual agency may require more collective coordination, more background governance, and sometimes more structured limits than libertarian intuitions prefer.

Practitioners can use this idea well. Many AI opportunities are really navigation problems, not content-generation problems. Many adoption barriers are really infrastructure and legitimacy issues, not model-quality issues alone. And many user benefits depend on shared systems that individual product teams cannot control by themselves. The misapplication would be to turn informational GPS into a fuzzy aspiration detached from context. Good guidance depends on good signals, bounded use cases, trusted pathways, and clarity about what counts as a better route. The book is right to elevate the metaphor. Waypoint has to do the harder work of specifying it.