Superagency

Picture your smartphone not as a distraction, but as a cognitive Ferrari sitting in your pocket — a machine that multiplies your ability to solve problems, navigate unfamiliar terrain, and shape your own life. What if the technology you've been told to fear is actually the exact tool you need to thrive?

That's the bet Reid Hoffman and Greg Beato make in this microbook. They argue we're standing at the edge of something they call superagency: a moment when billions of empowered minds, working alongside AI assistants, generate progress no previous generation could imagine. The catch is that this future won't arrive by accident. It depends on whether we steer toward it or freeze in panic.

Every major technology — from the printing press to the automobile — triggered apocalyptic forecasts before becoming an engine of freedom. AI is following the same arc, only faster. In the next minutes, you'll see why hands-on experimentation beats top-down fear, why "Big Knowledge" matters more than "Big Brother," and why your active participation is the missing ingredient in a future worth building.

The Techno-Humanist Compass

In the 15th century, doom-mongers warned that Gutenberg's printing press would unleash chaos, destroy memory, and corrupt the young. They were wrong in the most spectacular way: the press democratized knowledge and built modern civilization. Hoffman opens with that pattern because it keeps repeating. Telephones, automobiles, personal computers — each one provoked existential dread, then quietly expanded human agency.

When ChatGPT launched in late 2022, cultural anxiety pivoted overnight. Suddenly the fear wasn't social media addiction or platform monopolies — it was AI replacing human judgment altogether. Hoffman sorts the reactions into four camps: Doomers see extinction-level risk, Gloomers fixate on immediate social harms, Zoomers want to build with zero restraint, and Bloomers — the position he champions — treat innovation as agriculture, planting iteratively and harvesting gradual gains.

The Bloomer stance rests on what Hoffman calls a techno-humanist compass. We are Homo techne, a species that thinks through its tools. OpenAI's strategy of iterative deployment — releasing models, watching humans use them, fixing failures in public — isn't reckless. It's the safest path available, because it's the only one that learns in real time.

From Big Brother to Big Knowledge

In the 1960s, Vance Packard's bestseller The Naked Society warned Americans that a proposed Federal Data Center would create an Orwellian surveillance state. Critics imagined every citizen reduced to a punch card, every move tracked by faceless mainframes. The panic was so loud that the proposal died in Congress.

Sixty years later, we have far more data infrastructure than Packard ever feared — and far more individual freedom, not less. The digitization of information didn't deliver totalitarianism. It delivered self-publishing, remote work, global friendships, and platforms like LinkedIn, which Hoffman built specifically to turn personal data into scalable trust. A LinkedIn profile is voluntary data exchange that produces social capital: the ability to find work, mentors, and partners across continents.

What changed the script was framing. D. S. Halacy Jr.'s 1962 book Computers: The Machines We Think With saw the future correctly — these systems would amplify cognition, not replace it. Today we generate vastly more data than we can ever read. AI is the engine that transforms that passive flood into Big Knowledge: answers, summaries, and recommendations available the moment you need them.

The Cost of Problemism

Hoffman has a sharp name for the default Gloomer reflex: problemism. It's the habit of evaluating every innovation by what could go wrong, demanding perfect safety before deployment, and treating the status quo as risk-free. The problem with problemism is that the status quo is rarely safe.

Consider mental health. Hundreds of millions of people worldwide have no access to a therapist, and waitlists in wealthy countries stretch for months. That's the silent catastrophe. In 2023, Rob Morris ran a brief experiment on his peer-support platform Koko, layering GPT-3 into responses for users in distress. The backlash was immediate — critics called it unethical experimentation. But Kokobot, using techniques like cognitive reappraisal, was already helping real people who otherwise had nowhere to turn.

A JAMA Internal Medicine study compared ChatGPT's answers to those of human physicians on patient questions. Evaluators rated the AI's responses as more empathetic. Therapy has always been something of a black box — what works often defies measurement. Even without sentience, AI can deliver performative empathy at three in the morning when no human is available. Refusing that help in the name of theoretical harm isn't caution. It's a choice with a body count.

The Private Commons

Shoshana Zuboff's The Age of Surveillance Capitalism argues that Big Tech extracts value from users like oil from the ground. Hoffman offers a counter-frame: the private commons. Unlike a grazing pasture, digital information doesn't deplete when used. Every Waze driver reporting a pothole makes the map better for the next driver. Every Google search teaches the system to answer the next query faster.

The numbers tell a different story than the extraction narrative. Economists estimate the consumer surplus of search engines at around $17,530 per user per year — the dollar value people would demand to give up free search. Multiply that by billions, and the so-called extraction looks more like the largest wealth transfer to ordinary people in history.

Multimodal models now sit natively inside smartphones. Point your camera at a Spanish menu and it translates. Show it a rash and it suggests when to see a doctor. Ask it about a noise your car is making and it diagnoses the symptom. The private commons isn't taking from you. It's compounding for you.

The Evaluation Arena

The "AI arms race" headline suggests reckless labs racing past every guardrail. The reality inside AI development is closer to obsessive measurement. Every major model is graded on benchmarks like SuperGLUE, which test reasoning, comprehension, and bias across thousands of tasks. Competition pushes the numbers up — and pushes failure modes into the open.

Hallucinations and factual errors aren't permanent features. They're bugs being squeezed out through iteration, each release fed by data from millions of real conversations. The fix loop runs from the ground up: user surfaces a weird answer, lab patches it, next version is sharper.

Then there's Chatbot Arena, a crowdsourced platform where humans blind-test two model answers side by side and vote. It's direct democracy for AI capability — no marketing department, no cherry-picked demo. The leaderboard moves based on what real people actually prefer. That's not an arms race. That's accountability built into the product.

Permissionless Innovation and the Informational GPS

The precautionary principle says: prove it's safe before you ship it. It sounds responsible. In practice, it freezes progress and lets existing harms continue. Permissionless innovation — the American posture since the early 1990s internet — works the opposite way: ship, observe, adapt, regulate based on what you actually see.

The automobile is the textbook case. Early cars were lethal. William Phelps Eno wrote the first traffic rules in New York at the turn of the 20th century, not before cars existed, but after enough collisions revealed what rules were needed. Stop signs, lane markings, seat belts, airbags — each came from iteration, not preemptive bans. The same is true of GPS. In the 1990s, the U.S. ended Selective Availability, the policy that degraded civilian GPS signals. That single decision unlocked rideshare, logistics, precision agriculture, and trillions in downstream value.

AI is the next GPS — only for cognition. Hoffman calls it an informational GPS, a navigation system across the landscape of expertise. Prompt engineering unlocks what he calls latent expertise: the model already contains knowledge of marketing, law, code, biology. You just need to ask in a way that surfaces it. The biggest productivity gains in studies of AI deployment land on the least experienced workers. The junior employee with a model becomes nearly as capable as the senior one without.

Code as Law and Networked Autonomy

Lawrence Lessig warned in Code, and Other Laws of Cyberspace that software architecture quietly becomes the rule of law. Once a system is coded to forbid something, it forbids it absolutely — no judge, no appeal. We're entering that world physically now. The Driver Alcohol Detection System for Safety, or DADSS, would prevent a car from starting if it detects impairment. No debate, no second chance.

That perfect control cuts both ways. It saves lives, but it also strips out the human flexibility law has always allowed. Hoffman's answer is to make code more humane, not to reject it. AI-driven smart contracts can adjust to weather conditions, supply disruptions, or genuine hardship in ways paper contracts never could. Algorithmic equity, applied honestly, can route around the corruption and bias that have always tainted human enforcement. The frontier isn't whether code becomes law — it already is — but whether we build it with adaptability and consent baked in.

True autonomy, Hoffman argues, requires shared infrastructure. The Donner Party in 1846 had absolute freedom on the trail west, and that freedom killed them. Eisenhower's Interstate Highway System looked like a constraint — set lanes, fixed speeds — but it gave Americans real mobility for the first time. Networked autonomy beats raw individualism every time, and AI is the next layer of that network.

Building the Sovereign Future

In 1812, English Parliament passed the Frame Breaking Act, making it a capital offense for Luddites to smash mechanical looms. The rebellion failed, the looms won, and the regions that resisted automation lost their industries and their workers to places that embraced it. The lesson keeps repeating: brain drain follows technological refusal.

Nvidia's Jensen Huang has popularized the term Sovereign AI — the idea that nations must build their own models trained on their own languages and cultural data, or surrender that ground to foreign systems. France, India, the UAE, and Japan are all racing to do this. But Hoffman pushes the point further: governments themselves must adopt AI internally, automating citizen services to the standard of modern e-commerce. When renewing a passport feels worse than ordering from Amazon, trust in institutions corrodes. AI inside the state — handling permits, benefits, tax questions, healthcare navigation — is how democracies rebuild legitimacy.

Then there's the question of how citizens decide together. Taiwan's Polis platform uses machine learning to find consensus across thousands of participants, surfacing the statements most people actually agree with rather than amplifying the loudest fight. It has shaped real legislation on ride-sharing, alcohol sales, and platform regulation. This is the Copernican shift of the AI era — democracy upgraded from shouting match to consensus engine, with governance happening through participation, not surveillance.

Your Place in the Unmapped

The frontier of synthetic cognition has no map. Fear will tell you to wait for one. Don't. Safety emerges from steering, not from standing still — pick up the tools, experiment honestly, and claim your share in a world where intelligence becomes abundant, accessible, and unmistakably yours to wield.