o3,Gemini 2.5 and Beyond: Building Moats Amid AGI Uncertainty
The AGI debate is raging with no clear consensus on what constitutes artificial general intelligence. This article explores strategies to build lasting value in an unmeasurable AI era.
Multiple high-profile AI releases last week sparked an intense debate, particularly OpenAI's o3 model.
The argument about whether we've achieved artificial general intelligence (AGI) has grown heated, with George Mason University Economics Professor Tyler Cowen writing that April 16, 2025, should be considered "AGI Day." Meanwhile, research lab Transluce published findings showing significant flaws in these models, including a notable tendency to fabricate actions and justify them when confronted.
The AGI debate has raged for years, with no clear consensus on how to define it. Is it the ability to match human performance across all cognitive tasks? Or is it something more nuanced?
My stake in this debate is not to determine whether o3 constitutes AGI, but rather to explore what the potential arrival of AGI means for investors and companies navigating this technological Discontinuity. The absence of a clear definition or a timeline cannot be an excuse for inaction. Companies and investors that fail to prepare for AGI may find that hesitation fatal. Others will create a first-mover advantage.
The AGI Measurement Conundrum
Our ability to quantify artificial intelligence capabilities has reached a critical inflection point. Cowen probably overstated the case with his claim that "OpenAI's new release was 'AGI, seriously', singling out April 16, 2025, as some sort of historic 'AGI Day'."
This backlash over this proclamation exposes a fundamental truth: we lack reliable frameworks for measuring artificial intelligence in its most advanced forms.
At its core, intelligence involves the ability to understand, learn, reason, and apply knowledge to solve unfamiliar problems. While current AI systems demonstrate impressive pattern recognition and can generate content that appears intelligent, they fundamentally lack – at least for now - several hallmarks of human intelligence: self-awareness, intrinsic motivation, intuitive understanding of physical causality, and genuine cross-domain reasoning. AI can mimic these capabilities through statistical patterns, but does not possess the underlying cognitive architecture that humans use to navigate our complex world.
Beyond the definitional challenges, newer models like o3 demonstrate what Professor Ethan Mollick of Wharton calls the "jagged frontier" of AI capabilities. As Mollick explains, "the abilities of AI are uneven, even within a single job" where an AI might generate "great startup ideas but struggles to code complex products without human help." This uneven development creates a strategic landscape where competitive advantages can materialize or evaporate unpredictably.
These jagged capabilities are evident in recent testing results. According to TechCrunch, "o3 hallucinated in response to 33% of questions on PersonQA," which is "roughly double the hallucination rate of OpenAI's previous reasoning models."
In the same vein, Transluce found that o3 "frequently fabricates actions it never took, and then elaborately justifies these actions when confronted." In one particularly striking example, the model claimed to have run code on its own laptop "outside of ChatGPT" when challenged about its responses.
Similarly, Google's Gemini 2.5 shows its own pattern of strengths and weaknesses. Designed as "thinking models, capable of reasoning through their thoughts before responding," these models demonstrate impressive performance on benchmarks while still exhibiting limitations in real-world applications.
The progression of AI capabilities, while not yet reaching AGI, remains remarkable. Google's recent DolphinGemma project exemplifies this advancement, applying AI to the complex domain of inter-species communication. This AI model, built on Google's Gemma architecture, was "trained extensively on WDP's acoustic database of wild Atlantic spotted dolphins" to identify patterns and predict likely subsequent sounds in dolphin vocalizations. Despite having only 400 million parameters (modest by current standards), DolphinGemma demonstrates sophisticated capabilities in processing complex audio sequences and identifying patterns that would require immense manual effort by researchers. The model runs efficiently on Google Pixel phones, allowing field researchers to analyze dolphin communications in real time.
This project highlights both the remarkable progress and current limitations of AI systems. While powerful enough to identify patterns in complex non-human communication, the system still requires extensive human guidance and interpretation. It represents an impressive tool that augments human intelligence rather than a system that genuinely understands dolphin communication at a conceptual level.
The Single Point of Failure: AGI as Strategic Vulnerability
In this measurement vacuum, AGI—or even advanced narrow AI—represents a critical strategic vulnerability for organizations. Unlike previous technological disruptions that targeted specific industry verticals or business functions, advanced AI models like o3 and Gemini 2.5 represent a horizontal threat capable of simultaneously undermining multiple competitive advantages.
AI capabilities are advancing rapidly. The AI-2027 scenario, authored by a team including a former OpenAI researcher, presents a detailed progression that many experts find compelling. Their forecast shows a potential acceleration where "by late 2027, a major data center can hold tens of thousands of AI researchers that are each many times faster than the best human research engineer."
The agentic capabilities demonstrated by o3—its ability to decompose complex goals, use tools, and execute multi-step plans independently—suggest that synthetic networks can rapidly rival organic ones. When artificial intelligence can simulate, augment or potentially create network effects from scratch, the barriers to entry established by existing networks diminish rapidly.
Economies of scale have historically provided sustainable advantages through cost structures unachievable by smaller competitors. The latest generation of models fundamentally alters this equation by democratizing operational excellence. OpenAI claims its newest models can "independently use all ChatGPT tools – web browsing, Python, image understanding, and image generation" to "solve complex, multi-step problems more effectively and take real steps toward acting independently."
Perhaps more critical for businesses, the moats created through proprietary knowledge face accelerating depreciation. Both o3 and Gemini 2.5 demonstrate an unprecedented ability to reason across domains, apply contextual understanding, and generate novel solutions with minimal human guidance.
Gemini 2.5 models are designed as "thinking models, capable of reasoning through their thoughts before responding, resulting in enhanced performance and improved accuracy." This built-in reasoning capability means the half-life of proprietary knowledge advantages shortens dramatically. The competitive edge derived from specialized expertise becomes increasingly tenuous.
Building Durable Growth MoatsTM for the AGI Era
Faced with this unprecedented competitive landscape, forward-thinking organizations must construct fundamentally different types of moats—ones resistant to AGI-driven erosion and capable of sustaining growth even as discontinuity accelerates.
The material world offers perhaps the most durable moat against pure AGI disruption. Organizations with significant physical infrastructure—manufacturing facilities, logistics networks, and real estate portfolios—possess assets that remain beyond the direct reach of disembodied intelligence. While both o3 and Gemini 2.5 Pro excel at analyzing images and even creating visual simulations, they cannot directly manipulate physical reality. This limitation creates a strategic opportunity for material-world companies to systematically integrate AI capabilities while maintaining physical presence.
The most sustainable competitive positions may emerge not from resisting AGI but from pioneering new forms of human-AI collaboration. Organizations that develop proprietary frameworks for human-AI symbiosis—specific to their industry contexts and organizational cultures—create advantages that neither pure human nor pure AI systems can easily replicate. This symbiotic advantage extends beyond simple task allocation to encompass organizational structures, decision frameworks, cultural adaptations, and talent development systems that cultivate distinctly human capabilities complementary to AGI.
Perhaps counter-intuitively, the age of artificial general intelligence demands more distributed rather than more centralized organizational models. This distribution creates resilience against the single point of failure that AGI potentially represents. Decision rights allocated across human-AI systems, multiple parallel intelligence systems rather than monolithic deployments, redundant capabilities, and loosely coupled systems can limit cascade failures.
Strategic Imperatives Across the Business Landscape
The AGI Discontinuity presents distinct strategic challenges for different types of organizations.
For startups, the venture ecosystem faces fundamental recalibration as AGI capabilities advance. The most promising strategies involve building business models with AGI assumptions embedded from inception, developing organizational structures that naturally accommodate human-AI symbiosis, and focusing on creating tangible assets or relationship networks resistant to AGI replication. Ventures that position themselves as orchestrators of AGI capabilities rather than developers of proprietary algorithms may find more sustainable positions.
Tech incumbents face perhaps the greatest challenge in navigating the AGI Discontinuity. Their imperative includes conducting systematic audits of existing competitive advantages for AGI vulnerability, accelerating the development of human-AI collaboration frameworks, and restructuring organizational architectures to distribute intelligence rather than centralize it. Incumbents possess one significant advantage—existing customer relationships and data assets that provide the contextual foundation for AGI deployment. We will develop this concept further in the next weeks.
Enterprises with significant physical footprints—manufacturers, utilities, resource companies, real estate developers—occupy naturally defensible territory in the AGI landscape. Their strategic focus should include systematically integrating AGI capabilities into physical operations while maintaining ownership of material assets and developing proprietary interfaces between digital intelligence and physical systems.
Redefining Company Value in an AI World
The AI Discontinuity is also a challenge for investors. It fundamentally transforms how we must evaluate company value. Traditional valuation frameworks based on historical performance metrics become increasingly inadequate as AI systems transform how businesses operate and create value.
Valuation frameworks must evolve to account for the shift from “process automation” to “process autonomy”. Instead of applying multiples to current EBIT or revenue, investors must model which parts of the company can become agentic profit centers, how fast legacy costs can be replaced, and whether the company is positioned to own the intelligence layer in its ecosystem. This is, without doubt, the core question when assessing the strength of a company’s Durable Growth MoatTM.
Organizations with physical infrastructure moats must be valued not just on their current physical assets but also on their ability to integrate AI systems that transform these assets from capital-intensive liabilities into strategic anchors for autonomous operations. Similarly, companies building human-AI collaboration moats should be evaluated on their capacity to systematically transform knowledge workers into AI orchestrators who create value through frameworks competitors cannot easily replicate.
Investors who recognize this connection between moat building and valuation will not just find alpha, they will redefine what growth and margin expansion look like in an AI world. The first investors to develop reliable frameworks for identifying companies positioned to thrive in this Discontinuity will gain significant advantages in capital allocation and portfolio construction.
Navigating the Discontinuity Without Measurement
The fundamental challenge for both companies and investors remains: how to navigate a profound technological Discontinuity when we lack reliable ways to measure its progression or capabilities?
The solution requires a balanced approach combining Discontinuity-aware scenario planning, optionality as strategy, and strategic measurement evolution. Traditional scenario planning assumes relatively continuous development trajectories, but AGI requires approaches that incorporate capability threshold effects, model compressed timelines for competitive advantage erosion, and the development of contingency strategies for step-changes in AI capabilities.
In environments of radical uncertainty, optionality itself becomes strategic. Organizations should develop multiple parallel approaches to AGI integration, create reversible commitments that preserve strategic flexibility, and cultivate talent capable of navigating ambiguity and discontinuity.
While traditional benchmarks fail, organizations must develop new ways to measure AGI impact and capability by focusing on relative performance deltas rather than absolute capability claims and tracking organizational adaptability metrics rather than just technological metrics.
Conclusion: Decoding the Fundamental Discontinuity
The recent developments in o3, o4-mini, and Gemini 2.5 Pro—despite their flaws and limitations—signal a significant shift in how competitive advantage is created and sustained. While AI’s trajectory remains uncertain, its transformative impact is undeniable, rewarding those who prepare strategically.
The measurement vacuum at the heart of AGI development doesn't excuse strategic inaction—it demands more sophisticated approaches to building Durable Growth MoatsTM.
For investors and leaders alike, the winners will be those who see beyond traditional metrics to value resilience, adaptability, and intelligence-driven growth.