S
Reid Hoffman

Superagency

Entrepreneurship
Back to Categories
Entrepreneurship19 min read

Superagency

by Reid Hoffman

What Could Possibly Go Right with Our AI Future

Published: February 23, 2025
4.2 (12 ratings)

Book Summary

This is a comprehensive summary of Superagency by Reid Hoffman. The book explores what could possibly go right with our ai future.

what’s in it for me? learn how machines may make us mightier.#

Introduction

in the 15th century, religious authorities warned of a dangerous new invention – the printing press. the clergy warned it would unleash chaos, letting destabilizing ideas spread unchecked. a few centuries later, critics denounced the telephone, claiming it would replace genuine human connection with shallow exchanges. 

throughout history, breakthrough technologies have sparked fears of societal collapse, only to become essential to modern life. today, artificial intelligence (a.i.) faces similar skepticism. modern ai systems can engage in sophisticated conversations, solve complex problems, and even simulate human-like reasoning. while debates rage about what it all means, ai capabilities are remarkable enough to spark both wonder and worry.

at the heart of these worries lies human agency – our ability to maintain control over our lives and make autonomous choices. will ai eliminate jobs? will it erode privacy? will we become overly dependent on machines for tasks we once handled ourselves?

in this chapter, we will explore a more optimistic possibility. just as the industrial revolution used synthetic energy to dramatically expand human physical capabilities, ai represents an opportunity to amplify human intelligence and decision-making power. imagine every child having access to a personalized tutor as knowledgeable as leonardo da vinci, or a highly capable health advisor in their pocket. this combination of human and artificial intelligence could create what might be called “superagency” – an unprecedented expansion of individual human capability.

join us as we discover how ai might not just change our world, but enhance our human potential in ways we're only beginning to imagine.

mental health for the millions#

in early 2023, tech developer rob morris sparked an unexpected firestorm. his mental health messaging platform, koko, added ai capabilities to help compose supportive messages for users in emotional distress.  notably, users consistently rated these ai-assisted responses even higher than human ones. and every message clearly disclosed when it was written by an ai. even so, social media erupted with accusations of exploitation. 

the incident perfectly illustrates a tension in ai development: how do we balance solving urgent problems with managing hypothetical risks? 

the existing problems in mental healthcare couldn't be more urgent. in 2022, the united states recorded its highest suicide rate since 1941, with nearly 50,000 lives lost. another 100,000 people died from drug overdoses. to make matters worse, 129 million americans live in areas without sufficient mental health professionals, forcing many to wait three to six months for care. to put it plainly,  traditional solutions are falling devastatingly short.

current digital mental health solutions, including the over 10,000 available apps, show promise, but have significant limitations. up until recently, chatbots have relied on rigid, pre-programmed responses that can feel mechanical and impersonal. not surprisingly, only 3.9% of users continue with these apps after two weeks.

but what if mental healthcare could be delivered like spotify – accessible anytime, deeply personalized, and highly affordable? advanced ai systems could analyze millions of therapy interactions to understand what approaches work best for different people. unlike current chatbots, they could engage in nuanced, contextually-appropriate conversations that adapt to users' unique needs and circumstances. 

a 2023 study in jama internal medicine provided striking evidence of ai's potential: when physicians blindly evaluated medical advice from both human doctors and chatgpt, they rated the ai's responses higher in 78.6% of cases, finding them both more comprehensive and, ironically, more empathetic.

so, leaving your existing opinions on ai at the door, imagine this: a world in which everyone had access to as much clinically-validated therapeutic support as they wanted. users could test different therapeutic approaches, combine multiple styles, or assemble virtual therapy teams for real-time second opinions. researchers recently analyzed 160,000 anonymized therapy sessions containing over 20 million messages, using ai to identify which therapeutic approaches worked best in different contexts. this kind of data-driven insight could transform how we personalize mental healthcare.

the authors make clear that this vision is not about replacing human therapists. rather the goal is to provide more options for how mental healthcare can be delivered. ai could support human practitioners to serve more patients, and provide immediate support when human therapists aren't available. 

at the end of the day, this would mean a world with abundant mental healthcare. where regular access to mental healthcare through ai could reshape society’s capacity for kindness and understanding. it may sound far-fetched to some, but is it possible that – through neural networks and server clusters – we become not just more technologically advanced, but more humane?

a speed limit on progress#

in 1927, on a single day in new york city, eight children were killed in separate automobile accidents. critics called them “devil wagons,” while farmers dug trenches across roads to trap and disable these dangerous new machines. it’s a sentiment that feels remarkably familiar today as we grapple with artificial intelligence – another transformative technology that sparks both excitement and fear. 

but the story of automobiles actually teaches us something surprising about safety: innovation often proves more effective than regulation. rather than halting development, early car manufacturers and enthusiasts engaged in constant experimentation. they organized races and cross-country rallies that pushed the technology to its limits. while these events were risky, they led to crucial improvements in both reliability and safety. the results were remarkable: from 1923 to today, the death rate per 100 million miles driven has dropped by 93%.

meanwhile, this approach allowed cars to realize their revolutionary potential. imagine if, in 1923, strict regulators had limited cars to 25 mph. not only would we have forfeited the technology's transformative impact on individual freedom, economic opportunity, and social mobility, we may have also missed out on the safety improvements. 

this illustrates the power of what's called “permissionless innovation” – allowing new technologies to develop through rapid iteration and real-world testing. it stands in contrast to the so-called “precautionary principle,” which insists that new technologies must prove their safety before deployment. while this might sound prudent, it can actually make technologies less safe by slowing down the learning process – all while preventing or delaying their revolutionary benefits.

this lesson is particularly relevant for ai development. modern software allows for instant data collection and rapid updates, creating tight feedback loops that quickly reveal problems and enable solutions. when millions of people use ai systems daily, we gain valuable insights into real-world interactions, unexpected applications, and potential challenges across diverse contexts.

rather than trying to perfect ai systems behind closed doors, we should continue this process of iterative deployment – launching, learning, and improving. just as with automobiles, public engagement will help determine what risks we’re willing to accept based on the technology's value. if ai proves as transformative as the automobile in expanding human capability, the authors argue, we'll develop ways to manage its risks, while harvesting its benefits.

guiding progress with guardrails#

we’ve seen how knee-jerk over-regulation can stifle both innovation and safety. but that doesn’t mean all regulation is inherently restrictive or misplaced. can we instead use regulation to enhance, rather than hinder, human agency?

consider the 1800s, when settlers migrating westward faced a grueling journey. traveling by covered wagon, they moved at a painstaking pace – just 8 to 20 miles per day, depending on terrain, weather, and their own endurance. their paths were rough, often disappearing unexpectedly and forcing detours. reaching a destination could take six months or more, with winter’s arrival looming as a constant threat.

fast-forward to today: modern travelers can cover hundreds of miles in a single day with minimal planning, cruising in air-conditioned cars along smooth, well-marked roads. what once demanded months of hardship is now a leisurely road trip.

of course, driving today comes with regulations – speed limits, licensing requirements, seat belt laws. yet rather than restricting mobility, these measures have made it safer and more accessible. the interstate highway system is a prime example of how thoughtful regulation and infrastructure can expand individual freedom. originally conceived as a cold war defense project, it evolved into a vast network that transformed travel, making extraordinary mobility an everyday reality.

this brings us to artificial intelligence and the expansion of human capability through technology. like early automobiles, ai offers tremendous potential for expanding what individuals can accomplish. and like highways, making this power broadly accessible will require new frameworks of rules. 

just as driver's licenses and traffic laws made cars safer and more useful, ai may require new forms of certification and security measures. some worry this could restrict freedom – but consider how this might actually work: ai licensing could ensure safe access to powerful models while protecting against misuse. chains of data provenance could help prevent deep fakes. identity verification systems could allow widespread access to ai tools, while preventing impersonations and fraud. the right frameworks for ai deployment will help us to trust and adopt these systems, just as standardized road rules helped drivers confidently navigate america’s highways.

south korea's covid response provides another illustration of this principle. rather than implementing lockdowns, they enhanced individual freedom of movement by combining technology with clear regulations and public participation. the country used ai-powered contact tracing in a radically transparent system. this is how ai can expand human agency: through carefully designed systems that make those tools safely accessible to everyone.

the key is understanding that freedom isn't static – it evolves with our technological capabilities. new powers require new rules to become truly liberating.

onboard navigation for the information society#

remember life before gps – when paper maps weren’t "paper maps," but just maps? the evolution of gps offers a compelling parallel for understanding ai's potential.

initially, gps technology was reserved for military use. the first handheld consumer device didn’t appear until 1989 – and at a steep price, equivalent to $7,727 today. yet over time, gps became widely accessible, transforming the way billions of people navigate their daily lives.

ai is following a similar democratizing path; but instead of navigating physical space, it helps people traverse complex landscapes of information. consider a student with dyslexia who can now use ai to convert dense textbooks into accessible audio formats, weaving in examples from their personal hobbies or interests. or a recent immigrant who receives an intimidating legal notice in a non-native language; ai can explain the document's meaning in their own language and provide them with legal references they understand. ai responds to guidance and correction, allowing users to shape the interaction to their specific needs.

this democratizing effect is particularly powerful for those who previously lacked access to expertise. in a recent study, customer service representatives using ai assistance saw their productivity jump by 14% – but the most dramatic improvements came from newer employees, who could suddenly tap into years of expert knowledge on demand. for these workers, ai functions like a personal mentor, helping them quickly develop mastery of complex customer interactions.

unlike traditional sources of expertise, ai is instantly accessible, infinitely patient, and available 24/7. for a busy executive, this convenience saves time. but for someone juggling multiple jobs, limited transportation, and scarce childcare, it could be life-changing.

the key is that you direct the tool. just as you might tell your gps “avoid highways” or “find a route past that scenic overlook," you can guide ai to present information in ways that match your learning style, cultural background, and personal interests. it’s one thing to have access to information, theoretically; it’s another to navigate it confidently, on your own terms. 

upgrading democracy#

for years, the development of artificial intelligence was dominated by two nations: the united states and china. but today the ai landscape is becoming more multipolar, as nations recognize that commanding ai capabilities will be crucial to their future prosperity and autonomy. this has given rise to the concept of “sovereign ai” – the idea that countries must control their own ai development, to preserve both their strategic interests and cultural identity.

singapore has launched a national ai strategy specifically designed to reflect regional cultural values and norms. france has pledged $550 million to create its own “ai champions,” with president macron emphasizing the importance of developing french language databases. these programs reflect a growing understanding that ai will fundamentally shape how societies function and evolve.

this transformation is also helping governments reimagine public services. south korea is consolidating approximately 1,500 public services into a single ai-powered portal that will, for instance, proactively notify citizens about benefits they qualify for. the vision is to make government services as seamless as using amazon or google.

the implications for democratic governance are particularly fascinating. in taiwan, an ai-powered platform called polis has demonstrated how technology can enhance civic participation in unexpected ways. when taiwan faced a contentious debate over the regulation of ride-sharing apps, polis helped identify unexpected areas of consensus between pro- and anti-uber factions. the platform's design prevents trolling and angry exchanges, while using ai to cluster similar viewpoints and highlight areas of agreement. the result was a policy solution that earned broad public support. 

this points to a more optimistic vision of ai in governance – not as a tool for surveillance or control, but as a means to amplify citizens' voices in policymaking. forward-thinking countries are discovering that technology can strengthen rather than weaken democratic institutions, enabling more responsive and participatory governance. the nations that successfully navigate this transformation will find themselves with not just more efficient governments, but more engaged and empowered citizenries.

final summary#

Conclusion

the main takeaway of this chapter to superagency by reid hofman and greg beato is that artificial intelligence isn't just another technology to fear – it's potentially humanity's greatest tool for expanding individual capability. 

from democratizing mental healthcare and education to enhancing civic participation, ai could create “superagency” – a dramatic amplification of human potential. 

countless past technologies that once sparked similar anxieties have ultimately become essential to modern life. by embracing careful innovation rather than fear-based restriction, we can harness ai's transformative power while managing its risks, just as we did with books, automobiles, telephones, and gps.

okay, that’s it for this chapter. we hope you enjoyed it. if you can, please take the time to leave us a rating – we always appreciate your feedback. see you in the next chapter.