Agentic
Describes AI that acts with initiative: planning, choosing tools, and taking multiple steps toward a goal rather than replying once.
WHY IT MATTERS
It is the buzzword for the whole shift from chatbots to systems that get work done.
◈ THE FIELD MANUAL ◈
The plain-English AI glossary
Every term you keep running into, decoded in one plain read. No jargon walls. Each entry: what it means, why it matters, and where to go next. 53 terms and counting.
AI Agent
A program built on a language model that can take goals, make decisions, use tools, and act on its own across multiple steps, not just answer a single question.
WHY IT MATTERS
This is the shift from a chatbot that talks to a system that does. Agents can book, search, write, and update real systems.
Alignment
The ongoing work of making an AI's behavior match human intentions and values, so it is helpful, honest, and harmless.
WHY IT MATTERS
It is the difference between a capable model and a trustworthy one.
API (Application Programming Interface)
A defined way for one piece of software to talk to another. AI models are usually accessed through an API.
WHY IT MATTERS
It is how developers plug a model into apps, and how tools plug into an agent.
Attention
The mechanism inside a transformer that lets the model weigh how much each word should influence every other word when forming meaning.
WHY IT MATTERS
It is how a model keeps track of context and relationships across a sentence or a document.
Benchmark
A standard test used to compare models on a specific skill, such as reasoning, math, or coding.
WHY IT MATTERS
Benchmarks are useful signals but easy to over-trust. A high score does not guarantee a good fit for your task.
Chain-of-Thought
Prompting a model to work through its reasoning step by step before giving a final answer.
WHY IT MATTERS
It noticeably improves accuracy on math, logic, and multi-step problems.
Context Window
The maximum amount of text, measured in tokens, that a model can hold in mind at once, including your prompt and its own reply.
WHY IT MATTERS
Go past it and the model forgets the earliest parts. It sets how much document or history you can work with in one shot.
Diffusion Model
A type of model that generates images by starting from random noise and refining it step by step into a coherent picture.
WHY IT MATTERS
It is the technology behind most AI image generators, a different approach from the language models that power chat.
Distillation
Training a smaller model to imitate a larger one, capturing much of its ability in a cheaper, faster package.
WHY IT MATTERS
It is how providers ship small models that punch above their size for everyday tasks.
Embedding
A way of turning text into a list of numbers that captures its meaning, so a computer can measure how similar two pieces of text are.
WHY IT MATTERS
Embeddings power search, recommendations, and RAG. They are how AI finds "related" things by meaning, not keywords.
Few-Shot Prompting
Giving the model a handful of worked examples inside your prompt so it copies the pattern.
WHY IT MATTERS
Often the fastest way to lock in a format or style without any fine-tuning.
Fine-Tuning
Further training an existing model on your own examples so it adopts a specific style, format, or specialty.
WHY IT MATTERS
It bakes behavior into the model itself, useful when prompting alone cannot get consistent results.
Grounding
Tying an AI answer to real, checkable sources instead of the model's memory alone.
WHY IT MATTERS
Grounded answers can be trusted and cited. Ungrounded ones are just confident guesses.
Guardrails
Rules and filters that keep an AI inside safe, on-topic, and accurate boundaries, blocking bad outputs or risky actions.
WHY IT MATTERS
They are what makes an agent safe to point at real customers and real systems.
Hallucination
When a model states something false with full confidence, because it is predicting plausible-sounding language, not looking up facts.
WHY IT MATTERS
It is the number-one risk in real use. Anything factual from an AI needs verification.
Inference
The act of actually running a model to produce an answer. Training builds the model once; inference is every time you use it afterward.
WHY IT MATTERS
Inference is what you pay for per use and what determines speed. It is the running cost of AI.
Jailbreak
A crafted prompt that tricks a model into ignoring its safety rules and producing content it is meant to refuse.
WHY IT MATTERS
It is a core safety concern and the reason guardrails need constant testing.
Knowledge Cutoff
The date after which a model has no built-in knowledge, because its training data stopped there.
WHY IT MATTERS
It is why a model may not know recent events unless it is given search or fresh documents.
Latency
The delay between sending a request to a model and getting a response back.
WHY IT MATTERS
It shapes how an app feels. Reasoning models are more accurate but slower, a real trade-off to weigh.
LLM (Large Language Model)
A system trained on enormous amounts of text to predict the next chunk of language. That simple ability, at scale, produces reading, writing, reasoning, and answering.
WHY IT MATTERS
The LLM is the raw engine under every chatbot and agent. Everything else bolts onto it.
LoRA (Low-Rank Adaptation)
An efficient fine-tuning method that trains a small add-on layer instead of the whole model, saving time and compute.
WHY IT MATTERS
It makes customizing an open-weight model practical without the cost of full retraining.
MCP (Model Context Protocol)
An open standard for connecting AI models to external tools and data sources in a consistent way, so one tool works across many apps.
WHY IT MATTERS
It is becoming the universal plug for agents, the way a USB port standardized devices.
Memory
How an agent retains information: short-term memory for the current conversation, long-term memory for facts stored and recalled later.
WHY IT MATTERS
Memory is what lets an agent know you, your history, and what it did a minute ago.
Model
The trained file of learned patterns that an LLM runs on. When people say "which model are you using," they mean which specific trained brain.
WHY IT MATTERS
Different models have different strengths, costs, and limits. Picking the right one is a real decision.
Multi-Agent System
A setup where several specialized agents work together, each handling part of a task and passing results along.
WHY IT MATTERS
It is how bigger jobs get split up, one agent to plan, others to research, write, or check.
Multimodal
A model that can handle more than just text: images, audio, or video as input or output.
WHY IT MATTERS
It lets AI read a screenshot, describe a photo, or listen to a recording, not just chat.
Open Weights
A model whose trained parameters are released publicly, so anyone can download, run, and modify it on their own hardware.
WHY IT MATTERS
Open-weight models give you privacy, control, and no per-use fee, at the cost of running the infrastructure yourself.
Orchestration
The coordination layer that decides which model, tool, or agent handles each step of a workflow and in what order.
WHY IT MATTERS
It is the conductor of a multi-step AI system, turning separate pieces into one reliable process.
Parameters
The internal numbers a model learns during training, often in the billions. More parameters can mean more capability, but not always.
WHY IT MATTERS
Parameter count is a rough size measure people cite, though quality of training matters just as much.
Prompt
The instruction or question you give an AI. It can be one line or several pages of context, rules, and examples.
WHY IT MATTERS
The prompt is your steering wheel. Better prompts get better results from the exact same model.
Prompt Engineering
The craft of writing prompts that reliably get the result you want: clear instructions, context, examples, and format.
WHY IT MATTERS
It is the single highest-leverage AI skill for non-coders. Same model, far better output.
Prompt Injection
An attack where hidden instructions in a document, webpage, or message trick an AI into ignoring its real rules.
WHY IT MATTERS
It is a top security risk for any agent that reads outside content. Untrusted text can hijack behavior.
Prompt Template
A reusable, fill-in-the-blank prompt with slots for your specifics, so you get consistent results every time.
WHY IT MATTERS
Templates turn a good one-off prompt into a repeatable tool for a whole team.
Proprietary Model
A closed model you can only access as a hosted service through its maker, without seeing or downloading its weights.
WHY IT MATTERS
Most of the strongest assistants are proprietary. You trade control and privacy for convenience and power.
Quantization
Shrinking a model by storing its numbers at lower precision, so it uses less memory and runs faster, with a small quality trade-off.
WHY IT MATTERS
It is what lets large open-weight models run on ordinary computers instead of only on servers.
RAG (Retrieval-Augmented Generation)
A technique where the system first looks up relevant real documents, then hands them to the model so its answer is grounded in actual sources.
WHY IT MATTERS
It is the main defense against hallucination and how AI answers questions about your own private data.
Reasoning Model
A model tuned to spend extra effort thinking through a problem step by step before answering, trading speed for accuracy.
WHY IT MATTERS
It handles hard logic, math, and planning better, at higher cost and slower speed.
Reinforcement Learning
A training method where a model is rewarded for good outputs and penalized for bad ones, so it learns preferred behavior.
WHY IT MATTERS
It is a big reason modern assistants feel helpful and polite rather than raw and random.
RLHF (Reinforcement Learning from Human Feedback)
A training step where humans rate model outputs, and the model learns to prefer the answers people found helpful.
WHY IT MATTERS
It is a big reason modern assistants feel polite and useful rather than raw and erratic.
Semantic Search
Searching by meaning rather than exact keywords, so a query finds relevant results even when the words differ.
WHY IT MATTERS
It powers modern search and RAG, matching intent instead of just matching text.
Soul File
A plain-language file that defines an agent's identity: its name, voice, job, and rules. It is a friendly name for a rich system prompt.
WHY IT MATTERS
It is the moment an agent stops being a something and becomes a someone. Change it and the agent changes.
Streaming
Delivering a model's answer token by token as it is generated, rather than waiting for the whole reply to finish.
WHY IT MATTERS
It is why chat assistants appear to type in real time, which feels faster and more responsive.
Structured Output
Forcing a model to answer in a strict format, such as JSON, so software can read the result reliably instead of parsing free text.
WHY IT MATTERS
It is essential for connecting AI to other systems, where a predictable shape matters more than prose.
System Prompt
A hidden set of standing instructions that shapes how the AI behaves for an entire conversation: its role, tone, rules, and boundaries.
WHY IT MATTERS
It is where an agent gets its identity and guardrails. Change it and the assistant becomes a different assistant.
Temperature
A setting from roughly 0 to 1 that controls randomness. Low temperature gives focused, predictable answers; high gives varied, creative ones.
WHY IT MATTERS
It is a practical dial: near 0 for facts and code, higher for brainstorming and writing.
The Agent Loop
The cycle an agent runs: think, act, observe, repeat, adjusting each step until the goal is met.
WHY IT MATTERS
This simple loop is where all the apparent intelligence of an agent comes from.
Token
The unit an LLM reads and writes in. A token is roughly three-quarters of a word, so "learning" might be one or two tokens.
WHY IT MATTERS
Usage is billed and limited in tokens, and the context window is measured in them. Tokens are the currency of AI.
Tool Use (Function Calling)
The ability of a model to call external functions, like search, a calendar, or your database, and use the results.
WHY IT MATTERS
Tools are what turn talk into action. Without them, an agent can only describe; with them, it can do.
Training
The one-time, expensive process of building a model by exposing it to huge amounts of data until it learns patterns.
WHY IT MATTERS
Training happens rarely and costs a fortune; using the finished model (inference) is what you do daily.
Transformer
The neural-network design behind almost every modern language model. It reads a whole sequence at once and learns which parts relate to which.
WHY IT MATTERS
It is the architecture that made today's AI boom possible. The "T" in GPT stands for it.
Vector Database
A database built to store embeddings and quickly find the ones most similar to a query.
WHY IT MATTERS
It is the memory bank behind RAG and semantic search, letting an agent pull the right document from thousands.
Zero-Shot Prompting
Asking the model to do a task with no examples, relying purely on its instructions and training.
WHY IT MATTERS
It is the default way most people prompt. Knowing when to switch to few-shot is the skill.
Want the guided version? Start with what an AI agent really is, then explore the full learning hub.