AI is moving fast, and the vocabulary is moving faster. TechCrunch AI just published a living glossary aimed at people who’ve nodded along to terms like LLM, RAG, and RLHF without quite knowing what they mean. Here’s a practical guide based on that glossary, so you can read AI news without the imposter feeling.
Quick Start
By the end of this guide, you’ll understand the core terms shaping AI conversations right now: AGI, AI agents, API endpoints, chain-of-thought reasoning, coding agents, compute, deep learning, and diffusion. No prerequisites. Just five minutes and a willingness to learn what the headlines actually mean.
AGI (Artificial General Intelligence)
This one is fuzzy on purpose. Generally, AGI means AI that’s more capable than the average human at most tasks. According to TechCrunch AI, OpenAI CEO Sam Altman has called it the “equivalent of a median human that you could hire as a co-worker.” OpenAI’s charter frames it as “highly autonomous systems that outperform humans at most economically valuable work.” Google DeepMind says “AI that’s at least as capable as humans at most cognitive tasks.”
Why it matters: when companies talk about racing toward AGI, they’re not racing toward the same finish line. Definitions shape strategy, safety claims, and billions in funding.
AI Agent
An AI agent is a tool that uses AI to perform a series of tasks for you, beyond what a basic chatbot does. Filing expenses, booking a restaurant, writing and maintaining code. The category is messy because the infrastructure is still being built. Think of it as an autonomous system that can chain multiple AI models together to finish multistep work.
API Endpoints
Picture hidden buttons on the back of a piece of software that other programs can press. Developers use these to build integrations, like letting one app pull data from another. Most smart home devices and connected platforms have them. The twist: as AI agents get smarter, they can find and use these endpoints on their own. That opens powerful, sometimes unexpected automation.
Chain-of-Thought Reasoning
Some questions you answer instantly (taller, giraffe or cat?). Others need pen and paper (chickens and cows with 40 heads and 120 legs equals 20 of each). Chain-of-thought means the model breaks a problem into smaller intermediate steps before answering. It takes longer but produces better results, especially for logic and coding. Reasoning models are LLMs optimized for this through reinforcement learning.
Coding Agent
A more specific flavor of AI agent, applied to software development. Instead of suggesting code for a human to paste, a coding agent writes, tests, and debugs autonomously. It can work across entire codebases, spot bugs, run tests, and push fixes. TechCrunch AI describes it as hiring a fast intern who never sleeps. A human still reviews the work.
Compute
Compute is shorthand for the raw computational power AI models need to train and run. It usually refers to the hardware itself: GPUs, CPUs, TPUs, and the data center infrastructure beneath them. When you hear about compute shortages or compute spending, that’s the bedrock of the entire industry.
Deep Learning
A subset of machine learning built on multi-layered artificial neural networks loosely inspired by the human brain. Deep learning models find important features in data on their own, instead of needing engineers to define them. They learn from errors through repetition. The catch: they need millions of data points and take longer to train, so they cost more to build.
Diffusion
Diffusion powers many image, music, and text generators. Borrowed from physics, the system slowly destroys data by adding noise until nothing recognizable is left. Then it learns the reverse process, reconstructing structured output from noise. That’s how tools like image generators conjure a coherent picture from static.
Why This Matters
AI literacy is becoming table stakes for anyone working in tech, marketing, ops, or product. Knowing the difference between a chatbot and an agent, or between an LLM and a reasoning model, changes how you evaluate vendors, write specs, and spot hype.
Next Steps
Bookmark the full TechCrunch AI glossary since they update it as the field evolves. Pick one term per week and try it in a real workflow: spin up a coding agent on a small bug, prompt a reasoning model with chain-of-thought, or test an AI agent on a booking task. Reading definitions teaches you the words. Using the tools teaches you what they actually do.
Full glossary and ongoing updates available at the original TechCrunch AI source.