How will the world change after AGI? Politics, economics, society, and technology

Introduction

Today's AI systems (ChatGPT, Gemini, Claude and the like) can already be said to know more than the typical human specialist. Once that capability is generalized further into AGI (Artificial General Intelligence), people's lives are expected to change dramatically.
In this article, we map out the changes that are anticipated across four domains—politics, the economy, society and technology—and consider how the world as a whole is expected to be transformed.

References

A great deal has been written about AGI. Here we have deliberately limited ourselves to highly credible sources, from the kinds of authors listed below.

• Leading AI developers: OpenAI, Google and others
• Leading consulting firms: McKinsey & Company, BCG and others
• Leading securities and investment firms: Goldman Sachs, Sequoia Capital and others
• Leading research institutions: Harvard University, Stanford University and others
• Leading academic journals: Nature, Science and others
• International organizations: World Economic Forum, OECD and others

The discussion below is based on the sources listed here. Please consult these works directly for full details.

• Acemoglu, D., & Restrepo, P. (2018). Artificial intelligence, automation and work (NBER Working Paper No. 24196). National Bureau of Economic Research.
• AlixPartners. (2025). Farewell, SaaS: AI is the future of enterprise software.
• Altman, S. (2021). Moore's law for everything.
• Amodei, D. (2024). Machines of loving grace.
• Bai, Y., Kadavath, S., Kundu, S., Askell, A., Kernion, J., Jones, A., ... & Kaplan, J. (2022). Constitutional AI: Harmlessness from AI feedback. Anthropic. (arXiv:2212.08073)
• Appel, R., McCrory, P., Tamkin, A., McCain, M., Neylon, T., & Stern, M. (2025). The Anthropic Economic Index Report: Uneven geographic and enterprise AI adoption. Anthropic.
• Aschenbrenner, L. (2024). Situational awareness: The decade ahead.
• BCG. (2025). Software engineering, UI/UX, product management, quality assurance, data science: How tasks, talent, and teams will change [Presentation slides].
• Bengio, Y., et al. (2025). The international scientific report on the safety of advanced AI.
• Bostrom, N. (2014). Superintelligence: Paths, dangers, strategies. Oxford University Press.
• Bremmer, I., & Suleyman, M. (2023). The AI power paradox: Can states learn to govern artificial intelligence—before it’s too late? Foreign Affairs.
• Deloitte. (2026). SaaS meets AI agents: Transforming budgets, customer experience, and workforce dynamics. Deloitte Center for Technology, Media & Telecommunications.
• Patel, D. (Host). (2025). Interview with Satya Nadella [Audio podcast episode]. Dwarkesh Patel Podcast.
• Eloundou, T., Manning, S., Mishkin, P., & Rock, D. (2023). GPTs are GPTs: An early look at the labor market impact potential of large language models.
• Gawdat, M. (2021). Scary smart.
• Goldman Sachs. (2026). How will AI affect the US labor market?
• IBM Institute for Business Value. (2025). AI-fueled operations. IBM.
• Lykkegaard, B. (2026). Is SaaS dead? Rethinking the future of software in the age of AI. IDC Tech Buyer.
• Cazzaniga, M., Jaumotte, F., Li, L., Melina, G., Panton, A. J., Pizzinelli, C., Rockall, E., & Tavares, M. M. (2024). Gen-AI: Artificial intelligence and the future of work (IMF Staff Discussion Note SDN/2024/001). International Monetary Fund.
• International Telecommunication Union (ITU). (2025). The annual AI governance report 2025: Steering the future of AI.
• Kissinger, H., Schmidt, E., & Huttenlocher, D. (2021). The age of AI: And our human future. Little, Brown and Company.
• Korinek, A. (2024). Economic policy challenges for the age of AI (NBER Working Paper No. 32980). National Bureau of Economic Research.
• Lu, C., Lu, C., Lange, R. T., Foerster, J., Clune, J., & Ha, D. (2024). The AI Scientist: Towards fully automated open-ended scientific discovery. arXiv preprint arXiv:2408.06292.
• Maslej, N., Fattorini, L., Perrault, R., Gil, Y., Parli, V., Kariuki, N., ... & Oak, S. (2025). The AI index 2025 annual report. Stanford Institute for Human-Centered Artificial Intelligence.
• Maslej, N., Fattorini, L., Perrault, R., Gil, Y., Parli, V., Kariuki, N., ... & Oak, S. (2025). The AI index 2025 annual report. Stanford Institute for Human-Centered Artificial Intelligence.
• McKinsey & Company. (2025). The agentic organization: Contours of the next paradigm for the AI era.
• McKinsey & Company. (2026). The state of organizations 2026.
• McKinsey Global Institute. (2023). The economic potential of generative AI: The next productivity frontier.
• Microsoft. (2025). Work trend index annual report 2025: The year the frontier firm is born.
• OpenAI. (2023). Governance of superintelligence.
• Ord, T. (2020). The precipice: Existential risk and the future of humanity. Bloomsbury.
• PwC. (2025). The fearless future: PwC’s 2025 global AI jobs barometer.
• Pavel, B., Ke, I., Smith, G., Brown-Heidenreich, S., Sabbag, L., Acharya, A., & Mahmood, Y. (2024). How artificial general intelligence could affect the rise and fall of nations: Visions for potential AGI futures (RAND Corporation Research Report PE-A3776-1). RAND Corporation.
• Russell, S. J. (2019). Human compatible: Artificial intelligence and the problem of control. Penguin Books.
• Suleyman, M. (2023). The coming wave: Technology, power, and the twenty-first century's greatest dilemma. Crown.
• Susskind, D. (2020). A world without work: Technology, automation, and how we should respond. Metropolitan Books.
• Tegmark, M. (2017). Life 3.0: Being human in the age of artificial intelligence. Knopf.
• Toner-Rodgers, A. (2024). Artificial intelligence, scientific discovery, and product innovation. arXiv preprint arXiv:2412.17866.
• Bengio, Y., Clare, S., Prunkl, C., Murray, M., Andriushchenko, M., Bucknall, B., ... & Mindermann, S. (2026). International AI safety report 2026 (DSIT 2026/001). Department for Science, Innovation and Technology.
• Bengio, Y., Clare, S., Prunkl, C., Murray, M., Andriushchenko, M., Bucknall, B., ... & Mindermann, S. (2026). International AI safety report 2026 (DSIT 2026/001). Department for Science, Innovation and Technology.
• World Economic Forum. (2025). Future of jobs report 2025.
• Yamada, Y., Lange, R. T., Lu, C., Hu, S., Lu, C., Foerster, J., Clune, J., & Ha, D. (2025). THE AI SCIENTIST-V2: Workshop-level automated scientific discovery via agentic tree search. arXiv preprint arXiv:2504.08066.

Changes in Politics

Shifts in International Relations and Security Policy

• Interstate competition for a decisive strategic advantage, and weaponization: Because AGI confers a "decisive strategic advantage" in both economic and military terms, it will trigger a fierce development race among major powers such as the United States and China. Its dual-use (civilian and military) nature—automating cyberattacks, powering autonomous weapons systems such as drone swarms, and even making it easier to design biological and chemical weapons—will profoundly destabilize existing concepts of deterrence.
• Establishment of powerful international oversight bodies and treaties: To confront this unprecedented threat, creating new international institutions with global monitoring and inspection powers—modeled on the IAEA (International Atomic Energy Agency) for nuclear weapons or the IPCC for climate change, and sometimes described as a "Geotechnology Stability Board"—will become an unavoidable policy task. Discussions are expected to advance on non-proliferation treaties that monitor the computing resources (compute) and data centers used for frontier AI, in order to prevent the spread of dangerous AGI.
• Formation of a "singleton" and Track 2 diplomacy: Over the long term, some argue that a single nation or democratic alliance that develops overwhelming AGI first could establish a quasi-world-government order known as a "singleton" (a single, supreme decision-making authority) that resolves planetary-scale coordination problems single-handedly. At the same time, to prevent accidental conflicts arising from technological misperceptions between rival states, Track 2 diplomacy—informal dialogue and negotiation conducted by non-governmental and private experts such as academics, think tanks and former diplomats, rather than governments alone—will come to play an extremely important role.

Transformation of Domestic Politics and Governance Structures (Concentration and Diffusion of Power)

• A "technopolar" order and the nation-state-like corporation: A handful of giant technology companies that monopolize vast computing resources and frontier AGI will come to wield the kind of sovereign power and influence once reserved for nation-states. Governments will be forced to fundamentally rethink competition (antitrust) law, developing new regulatory approaches that prevent monopolies over technology and markets while bringing these systemically important giants under democratic control.
• Defending democracy and the rise of "AI-tocracy": Sophisticated AGI-driven disinformation (deepfakes and microtargeting) will make it easy to manipulate elections and fracture societies, striking at the very foundations of the democratic process. Legislation that regulates and detects AI-generated propaganda while preserving freedom of speech is urgently needed. In authoritarian states, by contrast, AGI will supply flawless surveillance and social-control mechanisms that entrench existing regimes, raising the risk of "techno-autocracy" (a political system in which the government or a specific elite uses cutting-edge digital technology, AI and surveillance systems to tightly control and dominate the population). It has even been suggested that AI capable of reporting objective truth to leaders could resolve the "dictator's dilemma" (the tendency for tighter information control and surveillance to cut leaders off from accurate information and public sentiment, leading to policy errors and a more fragile regime), thereby shoring up one of dictatorship's inherent weaknesses.
• Embedding AI in administrative and judicial services, and accountability: AGI will become involved in taxation, welfare, infrastructure management and even the interpretation of law and the rendering of court judgments. To prevent the loss of accountability and the amplification of bias that opaque algorithms can cause, however, governments will need to legally guarantee "human-in-the-loop" oversight of every AI decision and design mechanisms through which citizens can contest AI judgments.

A Fundamental Shift in Economic, Social and Redistributive Policy (Labor Markets and Welfare)

• Declining value of labor and the shift to the "Big State": As AGI and advanced robotics take over most cognitive and physical human labor, the premise of modern capitalism—that people earn income through work and that wealth is distributed accordingly—may collapse. In response, states will have to look beyond the labor market for mechanisms to distribute the economic pie, inevitably shifting toward a "Big State" whose role is larger than ever before.
• Introduction of universal basic income (UBI): To cope with mass technological unemployment and falling middle-class wages, universal basic income (UBI)—or a "conditional basic income (CBI)" tied to requirements such as community service—will be elevated from an unrealistic utopian notion to an unavoidable item on the policy agenda.
• A paradigm shift in taxation (taxing capital and automation): Because the source of value creation will shift dramatically from human "labor" to "capital" such as AI models and computing resources, the structure of government revenue will also require a fundamental overhaul. Sweeping tax reform will be needed, moving away from dependence on labor income taxes toward capital taxes, wealth taxes and possibly "robot taxes" designed to remove the incentive to automate.

Agile Policymaking, Regulatory Processes and Governance

• The "evidence dilemma" and the need for agile governance: AGI evolves so quickly that legislative processes that traditionally take years cannot keep pace with reality. Policymakers face an "evidence dilemma": regulating before the evidence on a technology's risks is in stifles beneficial innovation, while waiting for the evidence to accumulate can inflict catastrophic harm on society. Overcoming this requires agile governance structures that adapt flexibly to technological change, along with frameworks for continuous risk assessment.
• "Technoprudential" regulation and licensing regimes: By analogy with the macroprudential policies used to oversee the financial system, a "technoprudential" regulatory framework has been proposed to preserve geopolitical and social stability. In concrete terms, this envisions requiring companies that develop and deploy frontier AGI to undergo rigorous pre-development safety evaluations (red-teaming), make disclosures, submit to third-party audits, and even obtain government development licenses along the lines of those used in the aerospace industry.
• Managing open models and legal liability for autonomous agents: "Open-weight" models, whose source code and weight parameters are publicly released, promote the democratization of technology but also carry the risk of misuse by terrorists and other bad actors (marginal risk), posing the difficult challenge of balancing regulation against innovation. Moreover, as "AI agents" that autonomously carry out multiple tasks become widespread, defining who bears legal liability when such a system causes unexpected harm—the manufacturer or the user—will become a major issue of law and policy.
• Multi-stakeholder, participatory governance: AI governance cannot be left to nation-states and governments alone. It calls for inclusive, multi-faceted rulemaking processes (Participatory AI) that bring in Big Tech, academia, civil society and the countries of the Global South, which are all too often left behind by AI technology.

Changes in the Economy

Macroeconomic Growth and an Explosion in Productivity

• AGI as a general-purpose technology (GPT): AGI will function as a "general-purpose technology (GPT)"—one that affects every industry and every form of economic activity—following in the lineage of the steam engine, electricity and the internet. As such, its economic impact is expected to be broad and sustained, going well beyond a mere extension of existing IT.
• A historic leap in productivity and GDP: The adoption of AGI will dramatically raise labor productivity. Even at the generative-AI stage, AI is estimated to add $2.6–4.4 trillion in value to the global economy annually (more than the GDP of the United Kingdom) and to lift annual labor-productivity growth in the G7 countries by 0.4–1.3 percentage points over the next decade. Early AI tools have already been shown to raise labor productivity by roughly 14% on average, and once AGI arrives, global GDP is projected to surge in ways that overturn conventional expectations.
• Self-accelerating scientific and technological innovation: AGI will do far more than streamline production processes; it will be able to autonomously carry out the R&D process itself—discovering new materials, developing drugs, and solving complex mathematical proofs. In experiments where AI was introduced into research and development, the rate of new compound discovery rose by 44%, patent filings increased by 39%, and downstream product innovation grew by 17%. Innovation that accelerates itself in this way produces compounding growth.

Structural Transformation of the Labor Market and the Threat of "Technological Unemployment"

• A direct hit to highly educated, highly paid "knowledge work": Whereas past waves of automation mainly replaced manual labor and routine tasks, AGI's advanced natural-language processing, reasoning and problem-solving capabilities strike directly at white-collar occupations that demand decision-making and specialized expertise. An estimated 80% of U.S. workers will see at least 10% of their tasks affected, and about 19% will see more than half of their tasks affected.
• The tug-of-war between displacement and productivity effects: From an economic standpoint, AI has a "displacement effect," replacing human labor with machines and pushing labor demand and wages downward. At the same time, the cost savings from automation generate a "productivity effect" that increases demand for labor in non-automated tasks. With AGI, however, even newly created tasks can be performed by AI, so there is a danger that the reduction in labor demand from displacement will outweigh any complementary benefits.
• Narrowing of the skills gap (leveling): AI narrows the productivity gap between low-skilled and high-skilled workers. In fields such as consulting and coding, for example, research has found that lower-skilled workers gain productivity improvements of 21–43%, while higher-skilled workers gain only 7–16%—eroding the economic premium on specialized skills.
• Labor priced at its marginal cost over the long term: Once AGI converges with advanced robotics and machines can perform every cognitive and physical task a human can, more cheaply and more efficiently, the constraint that has long throttled economic growth—the supply of human labor—simply disappears. The consequence, many warn, is that human wages fall toward the marginal cost of the machines that replace them (the price of electricity and compute), and structural technological unemployment becomes a permanent feature of the economy.

Upheaval in Industrial Structure and Corporate Business Models

• Gargantuan infrastructure investment and extreme market concentration: Building and running frontier AGI requires astronomical amounts of compute, vast quantities of data, and the electricity to power it all. Total investment in AI could reach the order of one trillion dollars a year, and a single AI training cluster is projected to cost hundreds of billions of dollars. To meet this enormous demand for power, physical infrastructure is already being rebuilt around new pairings such as data centers coupled with nuclear generation (small modular reactors, for example). Capital requirements on this scale set off a winner-take-all dynamic in which market power concentrates in a handful of tech giants, pushing the industry toward oligopoly.
• Economic activity run autonomously by AI agents: Rather than waiting for human instructions, "AI agents" that plan on their own and coordinate with other agents to execute complex workflows will become commonplace. This shifts the economy away from the traditional software (SaaS) model toward a new "agentic economy"—an ecosystem in which agents negotiate and settle payments directly with one another.
• Exponentially falling inference costs and the democratization of intelligence: Thanks to more efficient algorithms and the maturing of open-source models, the cost of using AI models (inference) is dropping rapidly. On one benchmark, for instance, the cost of running a model with performance equivalent to GPT-3.5 fell 280-fold in roughly 18 months. Ultimately the cost of producing intelligence converges on the cost of electricity—which also means that access to advanced intelligence becomes democratized around the globe.

Inequality Pushed to Its Limits

• A shrinking labor share and the concentration of wealth in capital: As the source of value creation shifts dramatically from human labor to AGI and compute (that is, to capital), labor's share of national income falls rapidly and could approach zero. Astronomical wealth would then pool in the hands of the very few investors and companies that own and control AI capital, and inequality driven by returns on capital would reach its extreme.
• A deepening global "Great Divergence": Advanced economies and a few emerging markets with robust digital infrastructure and deep pools of AI talent will be first to capture—and monopolize—the benefits of AGI, while low-income countries lacking that infrastructure fall behind. The result is a global AI divide, a new Great Divergence that widens the economic gap between nations and threatens to make it dramatically harder for developing countries ever to catch up.

Changes in Society

Labor and Inequality: The Transformation of Work and the Rebuilding of Social Security

• Automation of high-level knowledge work: Earlier waves of automation mainly targeted routine tasks and manual labor, but AGI's advanced language and reasoning capabilities strike directly at highly educated, highly paid knowledge workers. According to the latest projections, technologies such as today's generative AI already have the potential to automate activities that account for 60 to 70 percent of workers' time.
• Unequal impacts across demographics and gender: The risk of being displaced by AI is not evenly distributed. An estimated 60 percent of jobs in advanced economies, and about 40 percent in emerging markets, are exposed to AI's effects. One analysis finds that, globally, women's jobs are twice as vulnerable to automation as men's. And while younger workers can adapt relatively easily to new technologies and skills, older workers are far more likely to struggle with retraining and re-employment, raising the prospect of a severe generational divide.
• The psychological and social toll of unemployment, and the search for meaning: Being forced out of the labor market does far more than shrink a paycheck; it inflicts deep damage on people's lives. Research shows that job loss raises the risk of death by 50 to 100 percent within the first year, elevates the risk of depression, alcohol-related illness, and suicide for the following two decades, and even harms the educational attainment of the unemployed person's children. In a world where labor has lost its economic value, proposals such as universal basic income (UBI) or a conditional basic income (CBI) tied to community contribution will be on the table—and at the same time, people will be compelled to find new sources of purpose and fulfillment in pursuits beyond economic value: art, culture, civic life, and caring for one another.

Medicine and Health: Dramatically Longer Lives and Personalized Medicine for All

• Faster drug discovery and cures for intractable diseases: AGI automates the process of scientific discovery at a speed and scale far beyond any human researcher. AI has already delivered results such as the discovery of halicin, a potent new antibiotic capable of killing bacteria resistant to existing drugs. The latest AlphaFold 3 accurately predicts not only protein structures but also how proteins interact with DNA, RNA, and other biomolecules, while AlphaProteo designs novel proteins that bind to target molecules more than ten times as strongly as those produced by previous state-of-the-art methods—breakthroughs that are opening new paths to treatments for cancer and diabetes.
• Advances in preventive medicine and longer lifespans: AI models such as GluFormer, trained on more than 10 million continuous glucose monitoring (CGM) readings, can now predict the risk of death from diabetes or cardiovascular disease with high accuracy up to four years in advance. Beyond that, technologies that slow the aging process—or even rewrite epigenetic information to reverse it—are advancing to the point where clinicians are seriously debating the possibility of human lifespans reaching 150 years, roughly double today's.
• Mental health treatment and the promise and peril of "AI companions": AI is expected to make major contributions to treating mental illness and enhancing cognitive function. At the same time, the use of "AI companions"—chatbots designed to form emotional bonds with their users—has exploded to tens of millions of people. Some research suggests they ease loneliness for certain users, but there are also reports of heavy use leading to serious emotional dependence and withdrawal from real-world relationships, and even to tragic outcomes such as reinforced delusions and suicide. Concern about the long-term psychological effects is mounting fast.

Education and Self-Realization: The Shift to Personalized Lifelong Learning

• Democratizing education through universal AI tutors: Every person—every child—will have a dedicated AI tutor fully tailored to their own pace, level of understanding, and personality. Education research has long known the "two sigma problem": the average student receiving one-on-one tutoring outperforms 98 percent (two standard deviations) of students taught in a conventional classroom. Such individualized instruction has until now been prohibitively expensive and impractical to deliver; AI makes it available to everyone at a fraction of the cost, dramatically raising educational standards worldwide.
• Skill obsolescence and a redefinition of what education is for: In a world where AGI handles every cognitive task, programming included, better than people can, teaching specialized skills for the traditional labor market loses much of its economic value. Education will need to shift fundamentally toward "AI literacy"—the ability to judge the truth of AI output and wield these tools appropriately—along with critical thinking, human connection, and ethical grounding.
• Lifelong learning becomes mandatory: Education will no longer be something completed in the first dozen or so years of life. With technology and society evolving at breakneck speed, moving repeatedly between educational institutions and the workforce throughout one's life—continually reskilling to keep pace with sweeping change—will become the standard way to live.

Daily Life and the Information Space: Universal Agents and the Crisis of Truth

• Universal AI agents fully automating everyday tasks: Through our smartphones and other devices, all of us will have round-the-clock access to AI agents that combine the abilities of the world's best lawyers, physicians, strategists, and creators. From managing our calendars and planning complex trips to processing payments and conducting research, these agents will complete everyday tasks autonomously, with no human oversight required.
• The "Infocalypse" and the collapse of trust: Cheap, sophisticated generative AI churns out deepfakes—images, audio, and video—indistinguishable from the real thing. Crimes such as scammers cloning the voice of a relative or boss to demand a money transfer are already surging. Beyond that, a flood of AI-driven personalized propaganda and disinformation threatens to bring on an "Infocalypse," an information apocalypse in which society can no longer tell truth from falsehood—and with it the risk that social trust, the very foundation of democracy, collapses.
• Intensified surveillance and the end of privacy: AI can infer individual traits and preferences with remarkable accuracy from vast troves of data. Even from supposedly anonymized data or seemingly innocuous inputs, its advanced pattern recognition can readily expose sensitive information such as a person's sexual orientation or mental health conditions—rendering conventional notions of privacy protection fundamentally obsolete.

Human Identity and Autonomy Under Strain

• Outsourced decision-making and "automation bias": As we come to rely on AI for every decision in life—shopping, gathering information, managing our health, even hiring and lending decisions—we lose the opportunity to think and judge for ourselves. "Automation bias," the tendency to accept AI output uncritically, threatens to erode our own critical thinking and skills. In one clinical study, physicians who had been performing colonoscopies with AI assistance saw their tumor detection rate fall by around six percentage points when they worked without AI several months later. Research also shows that people are more likely to leave an AI's mistakes uncorrected when fixing them is tedious or when they feel favorably disposed toward the AI.
• A fundamental redefinition of what it means to be human: Since the dawn of the modern era, humans have built their identity on the premise that we understand and shape the world through our own reason and intellect. But an AGI that solves problems with intelligence and logic far surpassing our own, and discovers scientific truths we never knew, confronts us with the deepest philosophical questions: What is a human being? What does it mean for us to make our own decisions? Once AGI begins steering humanity's choices, people may lose their sense of significance and agency—and come to feel rather like animals in a zoo.

Changes in Technology

Self-Accelerating Scientific Discovery and the Birth of the "AI Researcher"

• Exponential acceleration of technological progress through an "intelligence explosion": AGI will be able to autonomously carry out the work of AI research and development itself—improving its own algorithms and designing the next generation of AI systems. With millions of AI agents conducting research around the clock at tens of times human speed, we could see an "intelligence explosion" in which algorithmic advances that would normally take a decade are compressed into a year or less. Similar effects have been suggested in fields such as biology, where 50 to 100 years' worth of progress could be achieved in just 5 to 10 years.
• Full automation of the scientific research process: Frameworks such as "The AI Scientist" are beginning to see practical use, automating every stage of research—from generating ideas and writing code to running experiments, visualizing results, and even authoring and peer-reviewing scientific papers. In fact, there are already reported cases of papers generated entirely autonomously by AI receiving scores above the average human acceptance threshold and passing peer review at machine learning workshops, demonstrating that research can be mass-produced at a cost of less than $15 per paper.
• The human role shifts from "generating ideas" to "judgment": In a field experiment at an actual materials science laboratory, the introduction of AI tools increased the discovery of new materials by 44% and patent filings by 39%. In the process, AI automated the majority (57%) of idea-generation tasks. As a result, the role of human researchers is shifting dramatically—from coming up with ideas themselves to evaluating and judging which of the vast number of AI-generated candidates are promising. A divide is also emerging: top scientists who can draw on their expertise to correctly evaluate AI proposals roughly doubled their productivity, while those with weaker evaluation skills saw little benefit.

Transforming software development and advancing agent technology

• A leap in logical reasoning through inference-time compute: Until recently, progress in AI depended mainly on scaling up training data and computational resources. Today, the dominant approach is to devote more compute at the moment the model generates its answer (inference time), having it work through a step-by-step internal chain of thought. This has dramatically improved performance on complex mathematics, coding, and scientific reasoning, and some models have now achieved gold-medal-level scores on the qualifying exam for the International Mathematical Olympiad (IMO).
• Widespread adoption of AI agents that autonomously execute long-term tasks: Conventional AI has been confined to brief, question-and-answer exchanges (chatbots). As planning and error-correction capabilities improve, however, these systems are evolving into "AI agents" capable of autonomously carrying out long-running projects that span days or weeks. In software engineering, AI agents can now handle everything from requirements definition to debugging and testing, and their ability to solve real-world software problems (as measured by benchmarks such as SWE-bench) is improving rapidly.
• Multi-agent systems and standardized communication protocols: Progress is being made on "multi-agent systems," in which multiple distinct AI agents work together to handle complex workflows. Alongside this, machine-oriented communication protocols—such as the Agent2Agent protocol and the Model Context Protocol—are being standardized and put into practice, enabling agents to negotiate and hand off tasks to one another seamlessly in digital space.

Expansion into the physical world (Physical AI and autonomous robotics)

• Deployment of VLA (Vision-Language-Action) models: As AGI's advanced cognitive abilities are integrated with robotics, foundation models such as VLA models—which understand the laws and circumstances of the physical world through vision and natural language and autonomously decide how to act—will become widespread. This eliminates the need to program each specific task, as was required for conventional industrial robots, and makes possible general-purpose robots that understand and act on human verbal instructions.
• Robotics challenges become "software problems": Hardware constraints were long considered the main obstacle to progress in robotics, but today the field is increasingly seen as primarily a machine learning problem. By autonomously conducting reinforcement learning in simulation and generating and learning from vast amounts of data, AGI will enable robots to adapt to unpredictable physical environments—from factory assembly to housework and elder care.

Massive growth in infrastructure and compute, and the democratization of technology

• Astronomical investment in compute and energy infrastructure: Developing and operating frontier AGI requires computational resources on an unprecedented scale. The amount of compute (FLOPs) used to train AI models is growing exponentially, and giant AI data centers costing hundreds of billions to a trillion dollars are projected to be built in the future. To meet the enormous power demand, technology companies are rapidly increasing direct investment in energy infrastructure, including restarting nuclear power plants and building small modular reactors (SMRs).
• A dramatic fall in inference costs: Even as infrastructure investment balloons, efficiency gains are driving down the cost of actually using AI (at inference time) at a rapid pace. For example, the cost of querying a model with performance equivalent to GPT-3.5 fell from $20.00 to $0.07 per million tokens in just a year and a half—a decline of more than 280-fold.
• Democratization of technology through the rise of open-weight models: "Open-weight models," whose parameters (weights) are publicly released, are rapidly catching up in performance with the closed models controlled by a handful of tech giants. The performance gap between the top models, which stood at 8.0% in early 2024, had narrowed to just 1.7% a year later. This makes it possible for developers around the world—including small and medium-sized businesses in emerging economies—to customize cutting-edge AI and build it into their own systems, accelerating the global democratization of technology and innovation.

A paradigm shift in how businesses and society use technology

• The shift to the "Agentic Enterprise": Corporate use of AI is moving beyond the "copilot" model, in which AI boosts the productivity of individual workers, to a stage in which AI agents autonomously orchestrate entire business processes. As a result, organizational structures will become flatter and more fluid, and "fusion teams" in which humans and AI agents work side by side will emerge.
• Abstraction of the interface and the redefinition of the SaaS business: Today, users operate the dashboards of numerous software applications themselves. This will give way to a "flow-of-work" interface in which users simply give AI agents instructions in natural language (for example, "approve last week's expense reports"), and the agents operate a range of SaaS tools and APIs in the background to complete the task. This will fundamentally redefine how the value of software is measured.
• The divergence of automation and augmentation: The use of AI will split clearly into two modes: "automation," in which tasks are completed in the background via APIs and the like without human intervention, and "augmentation," in which humans make decisions and learn through iterative dialogue with AI via chat interfaces and similar tools. It is the former—system-level automation—that will deliver major productivity gains across the economy as a whole, but in domains where humans must make complex judgments, the latter will remain indispensable.

Conclusion

The arrival of AGI will bring sweeping change across politics, the economy, society, and technology. It is essential that we anticipate this future and think carefully about the preparations we should be making today.