The debate over whether "true" AGI will be a monolithic model or use external scaffolding is misguided. Our only existing proof of general intelligence—the human brain—is a complex, scaffolded system with specialized components. This suggests scaffolding is not a crutch for AI, but a natural feature of advanced intelligence.
The perceived need for a new "continual learning" architecture is overstated. Current models can already achieve this functionally by building their own tools and apps based on new information. This reframes the challenge from a fundamental research problem to a practical prompt engineering and application design issue.
The key to continual learning is not just a longer context window, but a new architecture with a spectrum of memory types. "Nested learning" proposes a model with different layers that update at different frequencies—from transient working memory to persistent core knowledge—mimicking how humans learn without catastrophic forgetting.
The gap between AI believers and skeptics isn't about who "gets it." It's driven by a psychological need for AI to be a normal, non-threatening technology. People grasp onto any argument that supports this view for their own peace of mind, career stability, or business model, making misinformation demand-driven.
The true threshold for AI becoming a disruptive, "non-normal" technology is when it can perform the new jobs that emerge from increased productivity. This breaks the historical cycle of human job reallocation, representing a fundamental economic shift distinct from past technological waves.
The distinction between a "model" and an "agent" is dissolving. Google's new Interactions API provides a single interface for both, signaling a future where flagship releases are complete systems out-of-the-box, capable of both simple queries and complex, long-running tasks, blurring the lines for developers and users.
A key design difference separates leading chatbots. ChatGPT consistently ends responses with prompts for further interaction, an engagement-maximizing strategy. In contrast, Claude may challenge a user's line of questioning or even end a conversation if it deems it unproductive, reflecting an alternative optimization metric centered on user well-being.
"Vibe coding" platforms, which allow users to create apps from natural language, pose a direct threat to the B2B SaaS market. For simple workflows, it is becoming faster for a team to build its own personalized app than to navigate the sales, procurement, and integration process for an existing SaaS product.
In high-stakes fields like pharma, AI's ability to generate more ideas (e.g., drug targets) is less valuable than its ability to aid in decision-making. Physical constraints on experimentation mean you can't test everything. The real need is for tools that help humans evaluate, prioritize, and gain conviction on a few key bets.
The disconnect between AI's superhuman benchmark scores and its limited economic impact exists because many benchmarks test esoteric problems. The Arc AGI prize instead focuses on tasks that are easy for humans, testing an AI's ability to learn new concepts from few examples—a better proxy for general, applicable intelligence.