Shane Legg suggests a two-phase test for "Minimal AGI." First, it must pass a broad suite of tasks that typical humans can do. Second, an adversarial team gets months to probe the AI, looking for any cognitive task a typical person can do that the AI cannot. If they fail to find one, the AI passes.

Standard AI benchmarks are an engineering tool for measuring performance. A more scientific approach, borrowed from cognitive psychology, uses targeted experiments. By designing problems where specific patterns of success and failure are diagnostic, researchers can uncover the underlying mechanisms and principles of an AI system, yielding deeper insights than a simple score.

AI agents have become proficient at following a pre-defined strategy to execute tasks. The next major frontier, and a significant bottleneck, is the ability to explore open-ended environments and generate novel strategies independently. This is the core capability that benchmarks like ARC AGI v3 are designed to test.

The test intentionally used a simple, conversational prompt one might give a colleague ("our blog is not good...make it better"). The models' varying success reveals that a key differentiator is the ability to interpret high-level intent and independently research best practices, rather than requiring meticulously detailed instructions.

An AI coding agent's performance is driven more by its "harness"—the system for prompting, tool access, and context management—than the underlying foundation model. This orchestration layer is where products create their unique value and where the most critical engineering work lies.

Moving away from abstract definitions, Sequoia Capital's Pat Grady and Sonia Huang propose a functional definition of AGI: the ability to figure things out. This involves combining baseline knowledge (pre-training) with reasoning and the capacity to iterate over long horizons to solve a problem without a predefined script, as seen in emerging coding agents.

Traditional AI benchmarks are seen as increasingly incremental and less interesting. The new frontier for evaluating a model's true capability lies in applied, complex tasks that mimic real-world interaction, such as building in Minecraft (MC Bench) or managing a simulated business (VendingBench), which are more revealing of raw intelligence.

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.

Shane Legg proposes "Minimal AGI" is achieved when an AI can perform the cognitive tasks a typical person can. It's not about matching Einstein, but about no longer failing at tasks we'd expect an average human to complete. This sets a more concrete and achievable initial benchmark for the field.