5
By Adriana Hoyos, Professor of AI Economics, Digital Ecosystems & Geopolitics, IE University
The last time humanity faced a real intelligence shock, it wasn’t a new gadget cycle—it was evolutionary. When a species becomes decisively smarter than the rest, it doesn’t merely compete better; it rewrites the rules of survival, coordination and power. Artificial general intelligence (AGI) is the first technology that plausibly moves us into that territory: It’s not just a better software product, but the moment humans stop being the smartest agents on Earth. Elon Musk set a hard date for the transition in his latest public conversation with Peter Diamandis1, saying, “I think we’ll hit AGI next year in 2026”, adding that by 2030, he was “confident” that AI would exceed the intelligence of all humans combined.
A business audience should be allergic to hype. So the right place to start is not philosophy but evidence. Today’s systems already beat humans at slices of professional work, and the slices are getting larger, faster than institutions can metabolize. Stanford University’s HAI’s “2025 AI Index”2 documents how quickly capability can jump on hard, real-world benchmarks. On the SWE-bench—designed to measure whether an AI can fix real bugs in real codebases—performance moved from 4.4 percent in 2023 to 71.7 percent in 2024. (SWE-bench is an industry-relevant software engineering benchmark designed to evaluate whether an AI model, or AI-coding agent, can perform practical development tasks on real codebases, rather than solving simplified, academic-style programming exercises.)
That kind of curve is why “AGI will arrive someday” has quietly turned into “AGI might already be here” in everything but name. In markets, the label matters less than the operational reality; if a tool can do what a role does, at the cost and speed that rewires unit economics, the transition begins immediately—long before regulators, standards bodies or even corporate job architectures agree on vocabulary.
This is also why reducing AGI to “a smarter chatbot” misses the point. Chat is just the interface. The real step-change is a general problem-solver that learns, reasons, transfers knowledge, plans and adapts across domains. Once you believe that it exists, or is close, the strategic question becomes brutally simple: What happens when cognition is no longer scarce? The answer should frame the coming shift less as a feature race and more as an intelligence-density one.
The intelligence-density potential is vastly greater than what we’re currently experiencing, since, according to experts such as Elon Musk, Dario Amodei3 and Ray Kurzweil4, humanity is off by two orders of magnitude on what’s achievable, as algorithmic progress can happen even on the same hardware. Algorithmic efficiency has been a major driver of progress foryears. OpenAI has published work5 specifically measuring how much less compute (computational power) is needed over time to reach the same performance levels.
Yet a meaningful divide exists among the most influential voices in AI over whether AGI has truly arrived and whether superintelligence is a realistic near-term prospect.
Yet a meaningful divide exists among the most influential voices in AI over whether AGI has truly arrived and whether superintelligence is a realistic near-term prospect. Yann LeCun remains one of the clearest skeptics, arguing that current systems are extraordinarily capable yet still far from the durable, general, world-modeling intelligence that AGI would require, and even farther from the kind of artificial superintelligence often invoked in public debates. By contrast, Jensen Huang has recently treated the AGI threshold as effectively reached, while Geoffrey Hinton has repeatedly warned that recent progress is no longer merely narrow or academic but increasingly consequential on a civilizational scale.
Meanwhile, figures such as Elon Musk and Alex Karp continue to describe AI as transformative and potentially destabilizing, yet they are often more careful in public about drawing a definitive line and declaring that AGI is already here. Recent speculation surrounding Anthropic’s apparent accidental leak of materials referencing Claude Mythos, described in reports as its most powerful model yet, has only sharpened that tension, because such episodes suggest that frontier capabilities may be advancing faster inside laboratories than public releases imply. Even without proving that AGI has arrived, the Mythos episode adds to the sense that the argument is no longer about distant theory, but about how close the industry may already be to systems that invite the comparison in earnest.
A concrete, non-theoretical example of AI already surpassing humans in a professional task is protein-structure prediction.Google DeepMind’s AlphaFold6 helped unlock protein folding at scale, and the public AlphaFold Protein Structure Database now provides open access to predictions for more than 200 million proteins. This is not a novelty demonstration or a short hype; it revolutionizes drug discovery while accelerating enzyme design and basic biology in ways that ripple through pharma pipelines, materials science and industrial chemistry.
This is the pattern to watch: Once AI becomes the best performer in a domain that feeds other domains, it doesn’t just automate labor; it compresses the time between idea, test and deployment, while exponentially augmenting the impact—beyond human capabilities.
The most underrated risk—and advantage—of modern AI is not that it can write, summarize or translate; it’s that it can tailor in unimagined ways. A Nature magazine report7 on AI research describes results showing that when given personal information about an opponent, ChatGPT could be more persuasive than humans.
That is what persuasion at the population scale looks like in practice—not one viral post, but millions of micro-arguments tuned to individuals, with no chatbots needed. For companies, that’s the holy grail of marketing efficiency—and the nightmare scenario for trust, elections, consumer protection and brand integrity.
The same capability that can nudge someone toward healthier behavior can also coax them toward misinformation, polarization and/or fraud. The business takeaway is uncomfortable: Reputational risk is about to become algorithmic, notoccasional.
So why does 2026 feel like an inflection point rather than another step on a smooth curve? Because multiple curves are crossing thresholds at once: adoption, capability and infrastructure.
So why does 2026 feel like an inflection point rather than another step on a smooth curve? Because multiple curves are crossing thresholds at once: adoption, capability and infrastructure. Microsoft Research’s AI Diffusion Report8 (January 2026) stated that roughly one in six people worldwide was using generative-AI tools; one in six is not early adopter, it’s mass market.
When a tool becomes the default interface, it stops being software and starts being a layer of reality—how people search, decide, learn, buy, apply for jobs and interpret events. When that happens at scale, the competitive moat shifts from having informationto having the best system for turning information into action.
Tech infrastructure is the other half of the story, and it’s where old mental models break. AI progress used to be limited by modelideas and data. Today, after just a couple of years, power and compute are increasingly the bottleneck, which is why AI founders, entrepreneurs and experts spend so much time talking about electricity, chips and training clusters in the same breath as AGI.
For instance, the unexpected and incredible growth of every single large tech company’s training clusters9, as well as industry leaders’ emphasis on power-hungry scaling, are perfect examples of what is at the top of today’s economy’s priorities. On the other hand, independent energy analysts are converging on the same conclusion: Data centers’ electricity demand is set to surge, with the International Energy Agency (IEA) publishing analysis on energy demand from AI10 and data centers as a central constraint. This matters because it means the AI race is not just a talent-and-model race; it’s also a capex (capital expenditure), energy procurement and geopolitical supply-chain contest.
Economics follows. Goldman Sachs Research11 estimated that generative AI could raise global GDP (gross domestic product) by about 7 percent over a 10-year period and lift productivity growth, with outcomes depending on levels of adoption and task automation. Even if reality lands below that, it’s still a multi-trillion-dollar shift in value creation. Andcrucially, it won’t be distributed evenly. The winners won’t just be AI companies; they’ll be organizations that re-architect themselves around machine-speed decision cycles, with shorter planning loops, faster product iterations and operational processes designed for AI agents rather than human handoffs.
This is where the jobs conversation becomes outdated. The framing AI will replace jobs assumes that the organization remains intact and swaps humans for software, one role at a time. However, AGI changes the unit of competition. If a single system can design a product, write and test code, generate compliant documentation, run campaigns, negotiate procurement and optimize pricing in near real time, then what gets disrupted is not a job—it’s the justification for the many layers of corporate structure.
That doesn’t mean every company becomes a one-person firm overnight. It means the cost of coordination collapses, and the premium shifts to strategy, distribution, brand trust, proprietary data and physical-world execution. The awkward truth is that the white-collar stack is unusually exposed, because it’s largely digital and rules-governed. The pace is already shocking even to insiders, with around a 10-times improvement per year—a compounding rate expected to continue.
The next transition—AGI to ASI (artificial superintelligence)—will be when the temperature changes. Once systems can improve themselves, the ceiling becomes less about human-engineering bandwidth and more about compute, energy and the boundaries of physics. What Elon Musk refers to as the intelligence density point12 is essentially a claim that we’re early in exploiting what hardware can already do and that algorithms are unlocking that headroom rapidly.
This aligns with the broader research narrative—such as the one published by OpenAI13—that algorithmic progress meaningfully reduces the compute required for a given capability level, effectively multiplying available intelligence. At that stage, capabilities that sound like magic in boardrooms—new-materials discovery, automated science, radically faster engineering cycles—stop being moonshots14 and start being quarterly advantages for whoever has the best systems and the best access to scarce inputs (power, chips, data, talent and permissive regulation).
If ASI is plausible, then the downstream implications extend beyond software into biology and medicine, because those are information problems wrapped in messy physical constraints. The moonshots discussion explicitly touches on health and diagnostics; recent reporting noted Musk describing uploading an MRI (magnetic resonance imaging) to Grok15 and comparing the results with those of medical professionals, treating consumer AI as a first-pass analytic layer.
Meanwhile, examples such as AI-driven biology are already reshaping research norms, much as the internet changed what was searchable in text. The business consequences are straightforward: Healthcare, pharma, insurance and longevity markets will face both extraordinary upsides and severe disruptions as diagnostics, discovery and personalization become cheaper and more accurate.
Finally, there’s the geopolitical-diffusion story, which includes the view that China is accelerating global diffusion by pushing open models. Major Chinese players have moved toward open-source releases as a competitive strategy, and Western observers increasingly worry that cost and accessibility will drive adoption in emerging markets.
Reuters reported that Baidu planned to open-source16 its next-generation ERNIE model and make its chatbot free, explicitly as competition heated up and rivals’ open models gained traction. For instance, Microsoft has highlighted how low-cost, open models can spread quickly17 across the Global South, widening strategic influence, while experts, such as Eric Schmidt, have warned that a bizarre outcome could be that the largest US models remain closed18 while leading Chinese models are open, giving China a geopolitical advantage in global adoption.
Put all of this together, and the central message becomes hard to avoid: This is not a cycle you can wait out. AGI is not a product launch; it’s an intelligence-regime change. The right executive posture is neither panic nor denial—it’s sober re-architecture. Assume your competitors will have access to near-expert cognitive labor at software speed. Assume your customerswill, too. Assume persuasion and trust will become computational. Assume your internal processes, governance, risk and compliance will be stress-tested by change rates they were never designed to handle. And then act accordingly, because whether Musk’s exact dates are right or wrong, the compounding curve to which he has pointed is already visible in the data.
ABOUT THE AUTHOR
Adriana Hoyos is a strategy consultant, board director, and academic with over 25 years of international experience. Her current work focuses on the economics of artificial intelligence, digital transformation, platform markets, business strategy, international development, the impact and applicability of technology ecosystems, public policy, and geopolitics.
Adriana serves as an independent board member and chair of the AI Taskforce on the Board of Directors of Nesma United Industries (Saudi Arabia). She is also a member of the expert panels of the Millennium Project (USA). She is a member of the Nasdaq Center for Board Excellence (USA) and the GCC Board Directors Institute (Dubai). She serves on the advisory board of SciTheWorld (Spain) and on the board of trustees of the Anar Foundation (Spain).
Most recently, she served as an independent board director and member of the ESG Committee at Sacyr. She served as CEO and board director of Women’s World Banking Co. She was also a board director of the U.S. Government Digital Economy Taskforce, Ryse Women-Inclusive Finance, Think Value, TechnoServe, Ashoka Co, Mentor, Blacksmith, and Conexión Colombia, among others.
Earlier in her career, she held corporate banking roles at Citibank. Adriana is an adjunct professor at IE University, where she teaches the economics of artificial intelligence, digital ecosystems, and geopolitics, and for more than a decade, she has also served as a senior fellow at Harvard University. She is a Google Scholar, serves as a tester at Google DeepMind AI Labs, collaborates academically with xAI, and contributes to the United Nations’ work on AI governance.
Her diplomatic experience includes representing Colombia as economic attaché to Spain, directing Plan Colombia for Europe, and representing the country at the United Nations. She has also served as a senior advisor in three U.S. presidential campaigns and held other high-level roles in public policy strategy. A frequent international speaker, she has participated in leading global forums including the World Economic Forum (Davos), the Clinton Global Initiative, and the United Nations Economic and Social Council, among others. She holds triple citizenship —United States, Spain, and Colombia, bringing a distinctly global geopolitical perspective and strong cross-cultural insight.
References
1 YouTube: “Elon Musk on AGI Timeline, US vs China, Job Markets, Clean Energy & Humanoid Robots | 220,” Peter H. Diamandis, January 6, 2026.
2 Stanford University Human-Centered Artificial Intelligence (HAI): “2025 AI Index Report/Technical Performance.”
3 Wikipedia: Dario Amodei.
4 Peter H. Diamandis: “EP #125 The Man Who Predicted AGI Decades Ago w/ Ray Kurzweil.”
5 OpenAI.
6 Google DeepMind: AlphaFold.
7 Nature: “AI is more persuasive than people in online debates,” Chris Simms, May 19, 2025.
8 Microsoft: “Global AI Adoption in 2025: A Widening Digital Divide,” January 2026.
9 Epoch AI: “AI training cluster sizes increased by more than 20x since 2016,” Robi Rahman, October 23, 2024.
10 International Energy Agency (IEA): “Energy demand from AI.”
11 Goldman Sachs Research: “Generative AI could raise global GDP by 7%,” April 5, 2023.
12 Next Big Future: “Elon Musk Expects True AGI in 2026-2027 and Superintelligence About 2030 and Believes in Antiaging Now,” Brian Wang, January 6, 2026.
13 OpenAI.
14 Moonshots.
15 Business Insider: “Elon Musk says he recently got an MRI and uploaded it to Grok,” Brent D. Griffiths, January 9, 2026.
16 Reuters: “China’s Baidu to make latest Ernie AI model open-source as competition heats up,” February 13, 2025.
17 Associated Press (AP) News: “DeepSeek’s AI gains traction in developing nations, Microsoft report says,” Chan Ho-Him and Matt O’Brien, January 8, 2026.
18 Yahoo Tech/Business Insider: “Eric Schmidt said he worries that most governments will use Chinese AI models,” Shubhangi Goel, November 11, 2025.