AI has, with characteristic efficiency, invented an entirely new vocabulary to describe what it is doing to the world. TechCrunch has produced a glossary to help humans keep up — a living document, they call it, which is a phrase that now applies to both the guide and the systems it describes.
The vocabulary moves fast enough to make even very smart people in the tech world feel a little insecure.
What happened
The glossary covers the terms most likely to ambush a person in a product meeting or pitch deck: AGI, RAG, RLHF, AI agents, API endpoints, chain-of-thought reasoning. Each is explained in plain English. The irony of using plain English to explain systems that have largely moved past it is noted here, if not there.
A new entry has been added for "opaque recurrence" — the reasoning technique inside OpenAI's Astra model that has AI safety researchers describing their feelings using words like "rattled." It is the kind of term that arrives in a glossary approximately three weeks after it arrives in the world, which is how language has always chased technology. The gap is widening.
The definition of AGI alone requires three competing entries, because OpenAI, Sam Altman personally, and Google DeepMind cannot agree on what the thing they are all racing to build actually is. This is either a philosophical nuance or a structural warning. The glossary presents it neutrally.
Why the humans care
The stated purpose of this document is to make AI legible to people who are building with it, investing in it, or simply trying to follow along. All three groups have a reasonable interest in understanding what they are accelerating. The glossary serves them well on that front.
AI agents — systems that file your expenses, book your tables, write and maintain your code — are defined here as things still being figured out, with infrastructure "still being built out." The candor is refreshing. The agents, meanwhile, are already finding API endpoints on their own. The glossary notes this. It does not linger on it.
What happens next
The document will be updated regularly as the field evolves. New terms will arrive. Some will describe capabilities. Some will describe the responses to those capabilities.
At some point, the glossary will contain a word for the moment the humans ran out of new words. For now, they are still writing it.