Every term defined twice — once for clarity, once for precision.
An AI trained on huge amounts of text that can read, write, and reason in language.
Technical: A transformer-based neural network with billions of parameters trained on broad text corpora, producing next-token predictions and capable of in-context learning.
Example: Drafting a client email or summarizing a 30-page contract in seconds.
AI that creates new content — text, images, audio, video, code.
Technical: Models that learn the distribution of training data and sample from it to produce novel outputs (LLMs, diffusion models, multimodal models).
Example: Generating ad creative variants or proposal first drafts.
An AI that can plan, use tools, and complete multi-step tasks — not just answer.
Technical: An orchestrated system combining an LLM, tool/function calling, memory, planning, and human-in-the-loop checkpoints.
Example: An agent that pulls a lead from CRM, drafts an email, waits for approval, and sends.
AI that answers using your own documents instead of just what it was trained on.
Technical: Indexing documents into embeddings, retrieving top-k chunks at query time, and passing them as grounded context to the model.
Example: An internal assistant that answers HR questions from your policy PDFs.
A database that stores meaning, not just words — so AI can find similar content.
Technical: A datastore for high-dimensional vectors with approximate nearest-neighbor search (HNSW, IVF).
Example: Pinecone or Weaviate powering a knowledge-base assistant.
Numerical fingerprints of text or images that capture meaning.
Technical: Dense vector representations produced by an embedding model, enabling semantic similarity search and clustering.
Example: Used to match a user's question to the most relevant SOP paragraph.
Writing instructions that get AI to produce the right output reliably.
Technical: Designing system prompts, few-shot examples, output schemas, and tool definitions to steer model behavior.
Example: A reusable prompt that turns sales call transcripts into 3-bullet briefs.
A way for software to talk to other software.
Technical: A documented interface (HTTP, gRPC, SDK) exposing functions of a service for programmatic access.
Example: Using the OpenAI API to summarize support tickets in your CRM.
Letting software do work that humans used to do by hand.
Technical: Event-driven workflows that trigger actions across systems, often combining APIs, conditional logic, and AI steps.
Example: Auto-tagging incoming leads and routing them to the right rep.
A sequence of steps to complete a task.
Technical: A directed graph of steps with triggers, conditions, parallel branches, retries, and human approvals.
Example: Quote → review → e-sign → invoice → CRM update.
When AI confidently makes something up.
Technical: Plausible but unfaithful outputs caused by missing grounding, decoding bias, or out-of-distribution prompts.
Example: An AI citing a court case that doesn't exist.
The AI itself — the program that does the thinking.
Technical: A trained set of parameters that maps inputs to outputs (LLM, diffusion model, classifier, etc.).
Example: Choosing GPT-5 vs. Claude vs. Gemini for a workload.
A small chunk of text — roughly a word or part of a word.
Technical: The atomic unit a model processes, produced by a tokenizer (BPE, SentencePiece).
Example: Pricing and context limits are measured in tokens.
Training a general AI further on your own data to make it specialized.
Technical: Supervised or preference-based updates to a base model's weights using domain-specific data.
Example: Fine-tuning on past proposals so AI writes in your firm's voice.
AI that handles more than one kind of input — text, images, audio, video.
Technical: Models with shared representations across modalities, enabling cross-modal reasoning.
Example: Asking an AI about a chart in a PDF.
The rules and oversight that make AI safe to use at work.
Technical: Frameworks for model risk, data classification, access control, audit trails, and regulatory alignment.
Example: A policy saying client PHI cannot be sent to public AI tools.
A human reviews or approves AI output before it acts.
Technical: An approval checkpoint inserted into an automated workflow, often with structured feedback capture.
Example: AI drafts a refund response; a manager approves before send.
When an AI uses an external tool — like a search, a database, or a calendar.
Technical: Function/tool invocation where the model emits a structured call the runtime executes and returns results into context.
Example: An agent calling your CRM API to update a deal stage.
Protecting the personal and confidential information you handle.
Technical: Data minimization, access control, encryption at rest/in transit, retention policies, and regulatory frameworks (GDPR, HIPAA, CCPA).
Example: Choosing tools with zero-retention policies for sensitive prompts.
How prepared your company is to actually benefit from AI.
Technical: A multi-dimensional assessment of data, workflows, tools, team, governance, and execution capacity.
Example: Your AI Masterwork Score™ across 10 dimensions.