Decoding AI's Expanding Vocabulary: Why Shared Definitions Matter

The Glossary as Industry Snapshot

According to TechCrunch AI, the field’s rapid evolution has generated a vocabulary gap: practitioners, investors, and media often deploy terms—LLMs, RAG, RLHF, AGI—without shared definitions. TechCrunch’s updated glossary treats this semantic fragmentation not as a temporary inconvenience but as a living document reflecting genuine disagreement on first principles. The publication’s approach reveals that terminology disputes are not merely linguistic; they encode different assumptions about what constitutes progress, capability, and risk in AI systems.

Artificial General Intelligence Remains Contested Ground

The glossary’s treatment of AGI crystallizes the problem. Artificial General Intelligence (AGI) lacks consensus even among its most prominent proponents. According to TechCrunch, OpenAI CEO Sam Altman frames AGI as “the equivalent of a median human that you could hire as a co-worker.” OpenAI’s formal charter advances a narrower definition: “highly autonomous systems that outperform humans at most economically valuable work.” Google DeepMind’s interpretation diverges further, describing AGI as systems “at least as capable as humans at most cognitive tasks.” TechCrunch notes that even experts researching AGI acknowledge confusion around these definitions. This definitional scatter has practical consequences—venture capital allocates funding, regulatory bodies draft policy, and product teams set timelines around competing interpretations of the same term.

Agents and API Endpoints: The Emerging Infrastructure Layer

TechCrunch distinguishes AI agents from simpler chatbots by emphasizing autonomy across multistep processes. An AI agent files expenses, books reservations, or maintains codebases without sequential human authorization. Critically, according to the glossary, agents increasingly locate and invoke API endpoints—the interfaces developers expose for programmatic control—without human intermediation. As agent capabilities expand, the ability to discover and call these hidden APIs introduces both automation potential and unpredictable failure modes. The publication notes that infrastructure for agentic systems remains “still being built out,” signaling that product definitions are hardening faster than the technical scaffolding supporting them.

Reasoning as Explicit Scaffolding

Chain-of-thought reasoning represents a simpler but instructive case: asking language models to articulate intermediate steps before answering. The glossary uses a farmer’s livestock math problem as its explanatory anchor—showing that decomposing complex tasks into substeps mirrors human problem-solving. TechCrunch frames this not as a novel discovery but as a technique that became central to modern LLM behavior only after explicit prompting research demonstrated its impact on accuracy.

Why This Matters

The glossary’s persistence as a “living document” signals a deeper reality: AI terminology will remain unstable as long as capabilities, deployment patterns, and stakeholder incentives continue shifting. Product teams building agentic systems cannot wait for industry consensus on agent definitions; they must ship with working prototypes and accept that the term will be redefined by what ships first. Researchers, regulators, and investors who rely on these terms—especially AGI—operate under the assumption of stability that the field does not provide. Organizations evaluating vendor claims around autonomy, intelligence, or generalization should interrogate definitions explicitly rather than accepting the glossary’s account as authoritative. TechCrunch’s framing of terminology as contested and evolving is itself newsworthy: it legitimizes skepticism toward confident claims about “what AI can do” by exposing the semantic ground beneath such claims as unresolved.