AI & Automation Glossary
A plain-English glossary of the terms you'll actually hear in 2026 — from AI agents and MCP to GEO and context engineering. Built for business owners, not data scientists.
132 terms
A/B Testing (with AI)
Running two versions of something — an email subject line, a landing page, a chatbot script — to see which performs better. AI-powered A/B testing goes further by automatically adjusting in real time based on results instead of waiting for you to check a dashboard and make the call.
A2A (Agent-to-Agent) Protocol
A standard for how one AI agent talks to another, even when they're built by different companies. Think of it as MCP for agent-to-agent communication. As multi-agent systems and "agentic commerce" pick up, A2A is the plumbing that lets your agent talk to a vendor's agent without custom integration work.
AEO (Answer Engine Optimization)
Structuring your content so AI assistants pull it as the direct answer when someone asks a question. Closely related to GEO and SEO — Google has publicly said AEO and GEO are still just SEO. The lines are blurring, but the term is everywhere right now.
Agent (AI Agent)
A piece of software that can take actions on its own based on a goal you give it. Instead of just answering a question, an agent can go book a meeting, send a follow-up email, look up a contact, or run a multi-step workflow. Think of it as a digital employee that follows instructions and adapts as it goes. Agents are the next evolution beyond simple chatbots — they don't just talk, they do.
Agent Experience (AX)
The agent equivalent of UX. How easy is your website, app, or product data for AI agents to read, navigate, and act on? As more customers send their AI to do research and make purchases on their behalf, AX is becoming a real design discipline — and a competitive edge for businesses that get it right early.
Agent Handoff
When one AI agent passes a conversation or task to another agent (or to a human) because it's hit the edge of what it can handle. Good handoffs are seamless — the next agent or person picks up with full context. Bad handoffs feel like being transferred to a new department and having to repeat everything.
Agent Loop
When an AI agent gets stuck repeating the same action or thought process without making progress. This happens when instructions aren't clear enough or the agent doesn't have a defined exit condition. It's the AI equivalent of an employee spinning their wheels. Good agent design includes guardrails to detect and break loops.
Agent Memory
The ability of an AI agent to remember information from previous interactions or tasks. Short-term memory covers the current conversation. Long-term memory lets an agent recall that a lead called last Tuesday and was interested in your premium package. Without memory, every interaction starts from zero.
Agent Orchestration
The system or logic that coordinates multiple AI agents working together. The orchestrator decides which agent handles what, when to pass tasks between agents, and how to resolve conflicts. Think of it as the manager in an AI team — it doesn't do the work itself but makes sure the right worker is on the right task.
Agent Persona
The personality, tone, and behavioral rules you give an AI agent. This includes how it greets people, how formal or casual it sounds, what it's allowed to say, and what it should refuse to do. A well-defined persona makes your AI agent feel like a natural extension of your brand instead of a generic robot.
Agent Team / Multi-Agent System
Multiple AI agents working together, each handling a different role or task. Instead of building one super-agent that does everything, you build a team of specialists. One agent qualifies leads, another books appointments, another handles support questions. They communicate and pass work to each other like a real team would.
Agent Tool Use
When an AI agent can use external tools — search the web, pull data from your CRM, send an email, update a spreadsheet — as part of completing a task. Tool use is what separates a basic chatbot from a real agent. The more tools an agent can access, the more useful work it can do independently.
Agent Washing
When vendors slap an "AI agent" sticker on what's really just basic automation or a glorified chatbot. Industry estimates suggest only a small fraction of "AI agent" products actually do agentic work. Worth knowing when you're vetting tools — ask what the agent can decide for itself, not just what it can execute.
Agentic AI
AI that doesn't just respond to questions — it actively takes initiative, plans multi-step tasks, and executes them with minimal hand-holding. "Agentic" is the buzzword for AI that behaves more like an employee than a search engine. When someone says "agentic workflow," they mean the AI is driving the process, not just assisting.
Agentic Commerce
When AI agents handle the buying. A customer sends their agent to compare products, negotiate, and complete the purchase on their behalf. On the B2B side, expect AI buyers negotiating with AI sellers in fractions of a second. Still early, but the infrastructure is being laid down right now.
Agentic RAG
A version of RAG (see Retrieval-Augmented Generation) where an AI agent actively decides what to look up, when, and from where — rather than pulling from one static knowledge base. More flexible and more accurate for complex questions where the answer lives across multiple sources.
Agentic Workflow
A process where AI agents handle most or all of the steps autonomously — making decisions, taking actions, checking results, and adjusting. Compared to traditional automation (rigid "if this, then that" rules), agentic workflows can adapt when something unexpected happens, much like a competent employee would.
AI (Artificial Intelligence)
Software that can do things that normally require a human brain — like reading emails, making decisions, writing content, or holding a conversation. When someone says "we use AI," they usually mean software that learns from data and handles tasks without being told every single step.
AI Audit
A systematic review of your business to find where AI and automation can save time, reduce errors, or make you more money. A good audit looks at your lead gen, sales process, delivery, ops, and admin — then gives you a prioritized list of what to fix first.
AI Citation
When an AI assistant references your brand or content in its answer. This is the new backlink — being cited by ChatGPT, Claude, or Perplexity now matters as much as ranking on page one of Google used to. Tracking AI citations is becoming a core marketing metric.
AI Governance
The policies, processes, and oversight your business puts around AI use. Who can use which tools, with what data, for what purposes, and how do you audit it? As AI moves into customer-facing and decision-making roles, governance stops being optional — especially with frameworks like the EU AI Act now in force.
AI Guardrails
Rules and limits you set on what an AI can and can't do. Examples: "Never quote a price," "Always transfer to a human if someone mentions legal issues," "Don't discuss competitors." Guardrails keep your AI from going off-script in ways that could hurt your business or brand.
AI Hallucination
When an AI confidently says something that's completely wrong or made up. It's not lying on purpose — it's pattern-matching and sometimes the pattern leads somewhere false. This is why you should never blindly trust AI output for facts, legal info, or anything high-stakes without checking.
AI Mode (Google)
Google's conversational AI search experience. Different from regular Google results — it gives a synthesized answer with cited sources. If your content gets cited in AI Mode, you show up in the answer even when nobody clicks your link. A key GEO target.
AI Optimization (AIO)
Another name floating around for GEO/AEO. Optimizing your content so AI assistants surface and recommend your business when answering customer questions. The industry hasn't settled on a single term yet — AIO, GEO, AEO, and LLMO are all in play.
AI Overview
The AI-generated summary that appears at the top of many Google searches. Pulls from multiple web sources and answers the question directly. If you're cited in an AI Overview, you get brand visibility even when users don't click through.
AI Receptionist
An AI agent (usually voice or chat) that handles front-desk duties — answering calls, greeting website visitors, routing inquiries, taking messages, and booking appointments. Works 24/7 and never calls in sick. This is one of the fastest-growing use cases for small businesses.
AI Slop
Low-quality AI-generated content flooding the internet — spammy blog posts, generic articles, mass-produced fake reviews. Search engines and AI assistants are getting better at filtering it out, and they're starting to penalize sources that produce it. A good reason to keep your content original and human-edited.
AI Stack
The combination of AI tools, platforms, and integrations you use in your business. Just like a "tech stack" refers to all your software, your AI stack is the collection of AI-powered tools working together. A well-designed AI stack has tools that complement each other rather than overlap.
API (Application Programming Interface)
The behind-the-scenes connection that lets two apps talk to each other. When your form submission automatically shows up in your CRM, that's an API at work. You don't need to understand how they're built — just know that APIs are what make integrations possible.
Automation
Any setup where work happens without you manually doing it. "If this happens, then do that." Could be as simple as auto-sending a welcome email when someone fills out a form, or as complex as a multi-step workflow that qualifies a lead, books a call, and updates your CRM.
Autonomous Agent
An AI agent that can operate independently for extended periods without needing human approval at every step. You set the goal and constraints, and it figures out the path. More autonomous = less babysitting, but also more risk if guardrails aren't set properly.
Batch Processing
Running a large number of tasks through AI all at once instead of one at a time. Example: taking 500 leads and having AI score, categorize, and write personalized outreach for each one overnight. Great for tasks that don't need real-time responses.
Bot / Chatbot
A program that holds a conversation with a person, usually through text on a website, SMS, or messaging app. Older chatbots followed rigid scripts. Modern AI chatbots can understand natural language, answer questions they haven't been specifically programmed for, and carry on a real conversation.
Browser Agent
An AI agent that controls a real web browser to complete tasks — shopping, booking, researching, filling forms. Examples include Claude in Chrome and similar tools from other vendors. Different from a computer-use agent in that it lives in your browser and works with the sites you already use.
Chain of Thought
A technique where AI is prompted to reason through a problem step by step before giving an answer, rather than jumping straight to a conclusion. Results in better, more accurate output — especially for complex tasks. When building AI agents, chain-of-thought prompting reduces errors.
Computer Use Agent
An AI agent that can actually control a computer — clicking buttons, filling out forms, navigating software — like a human would. Instead of needing an API integration, it literally uses the software the same way you would. Still early but evolving fast.
Constitutional AI
A training approach (pioneered by Anthropic) where the AI is taught a set of principles or a "constitution" that guides how it behaves — what to refuse, how to handle sensitive topics, how to stay helpful. It's part of why models like Claude behave the way they do.
Context Engineering
The next evolution beyond prompt engineering. Designing the full information environment around an AI — what it sees, what it remembers, what tools it has, when to retrieve information, and how to manage memory. If prompt engineering is writing a good email, context engineering is designing the whole communication workflow. Quickly becoming a core skill for anyone building serious AI systems.
Context Rot / Context Decay / Context Poisoning
What happens when too much irrelevant, outdated, or contradictory information piles up in an AI's context window. Performance drops, the AI gets distracted, or it starts repeating bad information. A reason "bigger context windows" isn't the whole answer — quality of context matters more than quantity.
Context Window
The amount of text (or data) an AI model can "see" and work with at one time. Think of it like the AI's working memory. A small context window means it forgets earlier parts of a long conversation. A large one means it can process an entire document or a long back-and-forth without losing the thread.
Conversational AI
AI that can hold natural, human-like conversations — not just answer single questions. This powers chatbots, voice agents, and AI assistants that can handle multi-turn discussions, remember what was said earlier, and ask clarifying questions. It's the technology that makes AI interactions feel less robotic.
CRM (Customer Relationship Management)
Software that tracks your leads, deals, and customer interactions in one place. Examples: HubSpot, Salesforce, GoHighLevel, Pipedrive. If AI and automation are the engine, your CRM is usually the dashboard.
Data Pipeline
The path your data takes from where it's created to where it's used. Example: a lead fills out a form → data goes to your CRM → gets enriched with AI → triggers an automation → lands on a report. A clean data pipeline means nothing gets lost or corrupted along the way.
Decision Tree
A structured set of if/then rules that guides an AI agent (or automation) through a process. "If the caller asks about pricing, say X. If they want to book, do Y. If they're angry, transfer to a human." Decision trees are the backbone of well-designed AI call agents and chatbots.
Deep Learning
A more advanced form of machine learning that uses layered neural networks to process complex patterns. It's what powers image recognition, voice synthesis, and the most capable language models. You don't need to understand the mechanics — just know it's the technology behind the most impressive AI capabilities.
Digital Clone
An AI version of you that can answer questions, share your expertise, and interact with people using your voice, knowledge, and style. Built by feeding your content (videos, writing, FAQs) into an AI model. Useful for coaching, onboarding, and scaling yourself without being in the room.
Distillation (Model Distillation)
Training a smaller, faster, cheaper model to mimic the behavior of a larger, more capable one. This is how you get close-to-frontier performance in a model that fits on a laptop or runs at a fraction of the cost. Why "small" models in 2026 are surprisingly capable.
E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness)
The signals search engines and AI assistants use to decide whether your content is worth citing. Anonymous content from no-name sources gets ignored. Named, credentialed authors with verifiable expertise win. Critical for GEO.
Embedding
A way of converting text (or images, or audio) into numerical representations that AI can compare and search through. This is the behind-the-scenes magic that lets AI find relevant information in your documents. When your AI chatbot pulls the right answer from your knowledge base, embeddings are doing the heavy lifting.
Escalation Path
The predefined route a conversation or task takes when an AI agent can't handle it. "If the customer mentions cancellation, route to the retention team." Good escalation paths mean your AI knows its limits and gets a human involved at the right moment — not too early (wasting human time) and not too late (frustrating the customer).
ETL (Extract, Transform, Load)
The process of pulling data from one place, cleaning or reformatting it, and putting it somewhere else. AI-powered ETL can handle messy data — like pulling info from unstructured emails or PDFs and organizing it into clean spreadsheet rows. Relevant when you're automating reporting or data entry.
Eval / Evals (Evaluations)
Tests you run to measure how well an AI system actually performs at your specific task. If you're not running evals, you're flying blind — "it felt good in testing" isn't a quality bar. Evals are becoming a core discipline for anyone running AI in production.
Extended Thinking / Reasoning Mode
A feature in newer AI models where the model "thinks" through a problem step by step before answering, using more compute for higher-quality results. Better for hard problems, more expensive per query. Sometimes billed as separate "reasoning tokens."
Few-Shot Prompting
Giving an AI a few examples of what you want before asking it to do the task. Instead of just saying "write a follow-up email," you show it two examples of good follow-up emails in your style, then ask it to write a new one. Dramatically improves output quality and consistency.
Fine-Tuning
Taking a general AI model and training it further on your specific data so it gets better at your particular use case. Like hiring a smart generalist and then giving them a month of onboarding so they understand your business, your tone, and your customers.
Foundation Model
A large, general-purpose AI model trained on broad data that can be adapted to a wide range of tasks. GPT, Claude, Gemini, and Llama are all foundation models. They're the "base layer" that most AI products are built on top of, either by prompting or fine-tuning.
Frontier Model
The most advanced, most capable AI models at any given moment — the ones pushing the limits of what's possible. Includes models like the latest Claude Opus, GPT, and Gemini Ultra. The label keeps moving as new models ship every few months.
Function Calling
The ability for an AI model to recognize when it needs to use an external tool and format the right request to do so. Example: you ask your AI "What's on my calendar tomorrow?" and it knows to call your calendar API rather than guess. Function calling is what makes agents actually useful rather than just conversational.
GEO (Generative Engine Optimization)
The practice of structuring your content so generative AI tools (ChatGPT, Perplexity, Gemini, Claude, Google AI Overviews) cite or recommend your business when answering questions. The AI-era equivalent of SEO — and increasingly more important, since AI answers are replacing clicks for a growing share of queries.
Generative AI
AI that creates new content — text, images, audio, video, code — rather than just analyzing or categorizing existing data. ChatGPT writing an email, Midjourney creating an image, and ElevenLabs generating a voiceover are all generative AI. Most of the AI tools relevant to your business fall into this category.
GPT (Generative Pre-trained Transformer)
The architecture behind ChatGPT and many modern AI tools. You don't need to know the technical details — just know that when someone says "GPT," they're usually referring to a large language model that can generate human-like text. It's become shorthand for "the AI that writes and talks."
Grounding
Connecting AI output to verified, factual sources rather than letting it generate from pure pattern-matching. A grounded AI agent pulls answers from your actual documents, your CRM data, or trusted databases. Reduces hallucinations and makes AI responses trustworthy enough for customer-facing use.
Human-in-the-Loop (HITL)
A system design where a human reviews, approves, or intervenes at key points in an AI workflow. Example: an AI drafts a proposal, but a human reviews it before it's sent. HITL is the smart middle ground between "fully manual" and "fully autonomous" while you're building trust in your AI systems.
Hyperautomation
Combining AI, automation, RPA (robotic process automation), and analytics to automate entire end-to-end processes — not just individual tasks. The strategy is to take a full workflow (lead-to-cash, hire-to-onboard) and remove every manual step that doesn't need to exist.
Inference
When an AI model processes an input and generates an output. Every time you ask ChatGPT a question, that's an inference. Relevant because some AI pricing is based on inference volume — the more you use it, the more you pay (usually measured in tokens).
Integration
Connecting two or more tools so data flows between them automatically. Example: connecting your website form to your CRM to your calendar to your email platform. Good integrations mean less copy-pasting and fewer things falling through the cracks.
Intent Recognition
An AI's ability to understand what someone actually wants, even if they don't say it perfectly. "I need to move my appointment" and "Can I reschedule?" mean the same thing. Good intent recognition is what makes AI voice agents and chatbots feel smart instead of frustrating.
Just-in-Time Retrieval
Pulling information into an AI's context only at the moment it's needed, instead of front-loading everything up front. Keeps context lean, costs lower, and answers sharper. A core context-engineering technique.
Knowledge Base
A structured collection of information that an AI can reference when answering questions. Could be your FAQs, SOPs, product docs, or training materials. The better your knowledge base, the smarter your AI tools become at handling real questions from real people.
KV Cache / Prompt Caching
A technique where the AI stores parts of a long prompt or document it's already processed, so it doesn't have to re-read everything on each interaction. Can dramatically reduce cost for systems that reuse the same context repeatedly — like a knowledge-base chatbot or an agent that re-reads your SOPs every turn.
Large Action Model (LAM)
An AI model designed not just to understand and generate text, but to take actions across software — clicking, filling forms, navigating apps. Closely related to agents, but the emphasis is on the model's ability to drive software directly.
Latency
The delay between when you send a request to an AI and when you get a response. In a chatbot, high latency means awkward pauses. In a voice agent, it means unnatural gaps in conversation. Low latency = the AI feels fast and responsive. This matters a lot for customer-facing AI.
Lead Scoring (AI-Powered)
Using AI to automatically rank your leads based on how likely they are to buy. Instead of manually reviewing every lead, AI analyzes behavior, demographics, and engagement patterns to tell you who's hot and who's not. Helps your sales efforts focus where they'll actually close.
LLM (Large Language Model)
The technical term for AI models like ChatGPT, Claude, and Gemini. They're trained on massive amounts of text and can generate, summarize, translate, and reason with language. Think of an LLM as the brain behind most of today's AI writing and conversation tools.
LLMO (Large Language Model Optimization)
Yet another name for GEO/AEO/AIO. Optimizing content for inclusion in LLM-generated answers. The terminology hasn't settled — pick one for your brand and stick with it.
llms.txt
A proposed standard file (similar to robots.txt) that tells AI crawlers what's on your site and how to use it. Adoption is uneven — Google has said it's not necessary for AI Overviews, but some marketers still recommend it for other AI tools. Worth knowing, optional to implement.
Machine Learning (ML)
The broader field of teaching computers to learn from data instead of being explicitly programmed for every scenario. AI is a subset of this. You don't need to understand the math — just know that "machine learning" is what lets AI tools get better over time as they see more data.
MCP (Model Context Protocol)
A standard that lets AI models connect to external tools, data sources, and services in a consistent way. Think of it as a universal adapter — instead of building custom connections for every tool, MCP gives AI agents a standardized way to plug into your CRM, calendar, databases, and other systems. Has become the de facto standard for tool use in 2026.
Mixture of Experts (MoE)
An AI architecture where the model is built from many specialized sub-models ("experts"), and only the relevant ones activate for each query. Makes models faster and cheaper to run without sacrificing capability. Many top 2026 models use this approach under the hood.
Model
The trained AI "brain" that powers a tool. When people say "which model are you using?" they're asking which AI engine is doing the thinking. Examples: GPT, Claude, Gemini, Llama. Different models have different strengths — some are better at writing, some at reasoning, some at speed.
Model Collapse
What happens when AI models are trained on too much AI-generated content. They start losing accuracy and diversity — essentially eating their own output and getting dumber over time. A real concern as more of the internet becomes AI-generated, and a reason high-quality original content is more valuable than ever.
Multi-Agent System
See "Agent Team." Multiple specialized AI agents working together on a shared objective. One might research, another might write, another might review. The power is in coordination — the system as a whole is smarter and more capable than any single agent.
Multi-Modal AI
AI that can process and generate more than one type of content — text, images, audio, video — within the same model or interaction. Example: you show an AI a photo of a whiteboard and it reads the text, interprets the diagram, and gives you a summary. Increasingly relevant as AI tools handle richer inputs and outputs.
Natural Language Processing (NLP)
The ability of software to understand and respond to normal human language — not code or commands. It's what lets you type "reschedule my Tuesday call to Thursday" into an AI assistant and have it actually understand what you mean.
No-Code / Low-Code
Platforms that let you build automations, apps, or workflows without writing traditional code. Examples: Make (formerly Integromat), Zapier, GoHighLevel, Bubble. These are the tools that let non-technical founders build real systems without hiring a developer for every change.
Onboarding Automation
A system that handles the steps of bringing a new client or customer on board — welcome emails, form fills, document collection, account setup, kickoff scheduling — without you doing each step manually.
Open Source AI
AI models and tools whose code is publicly available for anyone to use, modify, or self-host. Examples: Llama (Meta), Mistral, DeepSeek. Relevant because open source options can reduce costs and give you more control — but usually require more technical setup than commercial alternatives.
Outcome-Based Pricing / Per-Result Pricing
A pricing model where vendors charge per outcome (per qualified lead, per resolved support ticket, per booked appointment, per processed invoice) instead of per seat or per month. Becoming common with AI tools because the ROI is transparent — you can compare the cost of automation directly against what a human team would cost.
Parallel Agents
Multiple AI agents running simultaneously on different tasks rather than one after another. Example: while one agent researches a prospect, another drafts an email, and a third checks your calendar for availability. Parallel execution gets more done in less time.
Personalization Engine
A system that uses AI to automatically tailor content, offers, or experiences to individual users based on their behavior, preferences, or data. Example: an email sequence that adjusts its messaging based on which pages a lead visited on your site.
Prompt
The instruction or question you give to an AI. The quality of what you get back depends heavily on how you ask. "Write me an email" gives you generic output. "Write a 3-sentence follow-up email to a roofing company owner who attended my webinar but didn't book a call" gives you something useful.
Prompt Chaining
Connecting multiple prompts in sequence, where the output of one becomes the input for the next. Example: Prompt 1 researches a topic → Prompt 2 outlines an article based on that research → Prompt 3 writes the full draft. This is how you get complex, high-quality output from AI without trying to do everything in a single prompt.
Prompt Engineering
The skill of writing better prompts to get better AI output. It's less "engineering" and more "learning how to communicate clearly with the AI." Matters a lot when you're building AI agents, chatbots, or any system that relies on consistent AI responses.
Query Fan-Out
When an AI search system breaks one user question into several smaller searches behind the scenes to gather better information. If someone asks "best CRM for a 5-person roofing business," the AI might run separate searches for "best CRM for small business," "CRM for contractors," and "CRM pricing under $100." Relevant for GEO — you want to rank for the sub-queries too.
RAG (Retrieval-Augmented Generation)
A method where the AI pulls in specific information from your documents or data before generating a response — instead of just relying on what it was trained on. This is how you get an AI that actually knows your business, your products, and your policies instead of giving generic answers.
Rate Limiting
Caps on how many requests you can send to an AI service in a given time period. If you're running AI agents that make a lot of calls to an AI model, you might hit rate limits that slow things down. Relevant when scaling AI automations — you need to design around these limits.
Reasoning (AI Reasoning)
The ability of an AI model to think through problems logically rather than just pattern-matching. Newer models are specifically built for better reasoning — breaking complex problems into steps, considering multiple options, and checking their own work. Better reasoning = more reliable agents.
Reasoning Model
An AI model specifically built to think through problems step by step before answering, rather than just pattern-matching to a quick response. Better for math, code, planning, and complex multi-step tasks. Slower and more expensive per query, but significantly more accurate when accuracy matters.
Reflection
The ability of an AI agent to self-critique its own work, notice mistakes, and adjust its plan. The "wait, that didn't work — let me try a different approach" capability. A core piece of what makes an agent feel competent vs. brittle.
Retrieval
The process of finding and pulling relevant information from a database, document set, or knowledge base. When your AI customer support agent looks up the right answer from your help docs, that's retrieval. The quality of retrieval directly determines the quality of AI responses in RAG systems.
RLHF (Reinforcement Learning from Human Feedback)
A training technique where humans rate AI outputs, and the model learns to prefer the kinds of answers humans like. It's a big part of why modern AI sounds helpful and aligned instead of robotic or weird.
SaaS (Software as a Service)
Any software you pay for on a subscription and access through the internet. Your CRM, email marketing platform, project management tool — those are all SaaS. Relevant because most AI and automation systems are built by connecting multiple SaaS tools together.
Semantic Search
Search that understands meaning, not just keywords. Traditional search for "how to cancel" only finds pages with those exact words. Semantic search also finds pages about "ending my subscription" or "opt out of renewal" because it understands they mean the same thing. Powers better AI chatbots and knowledge base lookups.
Sentiment Analysis
AI that detects the emotional tone of text — positive, negative, frustrated, confused, urgent. Useful for automatically flagging unhappy customers, prioritizing support tickets, or adjusting how an AI agent responds based on how the person seems to be feeling.
Sequential Agents
AI agents that work one after another in a defined order, each building on the output of the previous one. Agent 1 researches → Agent 2 summarizes → Agent 3 drafts the email. The opposite of parallel agents. Better for tasks where each step depends on the one before it.
Shadow AI
When employees use AI tools (their personal ChatGPT account, a side subscription to Claude) for work tasks without company approval. A growing security and compliance headache — sensitive data ends up in random tools and nobody knows where it went.
Skill (Agent Skill)
A self-contained set of instructions and resources you give an AI agent to teach it how to do a specific task well. Modular, reusable, and easier to maintain than stuffing everything into one giant system prompt. Newer terminology (popularized by Anthropic in 2025) that's now industry-standard.
Small Language Model (SLM)
A smaller, more efficient version of an LLM. Less capable than the giants, but cheap enough to run on a laptop or phone, fast enough for real-time use, and private enough to keep data in-house. Great for specific, well-defined tasks where you don't need a frontier model.
SOPs (Standard Operating Procedures)
Step-by-step documentation of how recurring tasks should be done. SOPs are the foundation for good automation — you can't automate what you haven't defined. If your process lives only in someone's head, it can't become a system.
Structured Output
Getting AI to return data in a consistent, predictable format (like JSON, a table, or specific fields) rather than free-form text. Critical for building reliable automations — when your AI agent qualifies a lead, you need the result in a structured format your CRM can actually use, not a paragraph of prose.
Subagent
A specialized agent spawned by a parent agent to handle a focused subtask. The parent gives it a job, gets the result, and moves on. Standard terminology now in multi-agent architectures — and a useful concept when you're thinking about how to break a workflow into manageable pieces.
Supervisor Agent
An AI agent whose job is to manage other agents. It assigns tasks, reviews output, handles errors, and decides when to escalate to a human. In an agent team, the supervisor is the team lead — it doesn't do the grunt work but makes sure the team delivers the right result.
Swarm (Agent Swarm)
A large group of AI agents working together dynamically, often without a rigid hierarchy. Instead of a fixed team with defined roles, a swarm can self-organize based on the task. Still mostly experimental, but the concept is relevant as agent systems get more sophisticated.
Synthetic Data
Data generated by AI to train other AI models — often used when real data is scarce, sensitive, or biased. Useful, but overdo it and you get model collapse (see above). A real tool in the toolbox, not a free pass.
System Prompt
The hidden instructions that define how an AI behaves before a user ever interacts with it. When you set up an AI chatbot and tell it "You are a helpful scheduling assistant for a dental practice. Never discuss pricing. Always be friendly," that's the system prompt. It's the foundation of every AI agent's behavior.
Task Agent
An AI agent designed to complete one specific task and nothing else. "Check this inbox for new leads and add them to the CRM." Simple, focused, reliable. Task agents are the building blocks of larger agent teams — each one does one thing well.
Temperature
A setting that controls how creative or random an AI's responses are. Low temperature = predictable, consistent, safe. High temperature = creative, varied, sometimes surprising. For business operations and customer-facing agents, you usually want low temperature. For brainstorming and content creation, higher can be useful.
Test-Time Compute
The idea that you can get better AI answers by letting the model "think harder" at the moment you ask it (more reasoning steps, more compute), rather than only investing in bigger training runs. Why some models cost more per query but give noticeably better answers on hard problems.
Token
The unit AI models use to measure text. Roughly, 1 token ≈ ¾ of a word. Matters because most AI tools charge by token usage or have token limits on how much text they can process at once. If an AI tool cuts you off mid-conversation, you've probably hit a token limit.
Training Data
The information an AI model was taught on. This determines what it knows, how it thinks, and what biases it might have. A model trained mostly on English text will be weaker in other languages. A model trained on general internet data won't know the specifics of your business — which is why fine-tuning and RAG exist.
Trigger
The event that kicks off an automation or agent workflow. "When a form is submitted," "When a call comes in," "When a deal stage changes." Every automation starts with a trigger. Getting your triggers right is half the battle of building systems that actually work.
Vector Database
A specialized database designed to store and search embeddings (numerical representations of text, images, etc.). This is the technology that powers fast, accurate semantic search in AI applications. When your AI chatbot instantly finds the right answer from thousands of documents, a vector database is usually behind it.
Vibe Coding
A casual term (coined by Andrej Karpathy in early 2025) for the practice of describing what you want an app to do in natural language and letting AI write the code. You don't really read the code — you just see if it works and tweak the description. Has powered a wave of non-developers shipping real software.
Voice Agent / AI Voice Agent
An AI that handles phone calls — answering, qualifying, booking appointments, providing information — without a human on the line. It sounds natural, follows a script or logic you define, and works 24/7. This is the technology behind always-on AI sales and reception systems.
Voice Cloning
Creating a synthetic version of someone's voice using AI. Record a sample, and the AI can generate new speech that sounds like that person. Used for digital clones, personalized voice messages at scale, and content creation. Quality has gotten remarkably good — most listeners can't tell the difference.
Warm Handoff
When an AI agent transfers a conversation to a human and includes all the context — what was discussed, what the person needs, what's already been tried. The human picks up the conversation seamlessly. The opposite of a cold transfer where the customer has to start over from scratch.
Webhook
A way for one app to instantly notify another app when something happens. Example: someone fills out your form → webhook fires → your CRM creates a new contact → automation sends a welcome email. Webhooks are the trigger mechanism behind most real-time automations.
Workflow
A defined sequence of steps that move a task from start to finish. In automation terms, it's the chain of actions that happen when a trigger fires. Example workflow: new lead comes in → AI qualifies → CRM updated → calendar invite sent → reminder scheduled.
Workflow Automation vs. Agentic Automation
Traditional workflow automation follows fixed rules — "if this, then that." It's reliable but rigid. Agentic automation uses AI agents that can reason, adapt, and make judgment calls within boundaries you set. Most good business systems use a mix of both: rigid automation for predictable steps, agents for the parts that need flexibility.
Zapier / Make / n8n
Popular automation platforms that connect apps and build workflows without code. Zapier is the most well-known and beginner-friendly. Make (formerly Integromat) offers more complex logic. n8n is open-source and self-hosted. These are the "plumbing" behind most business automations.
Zero-Shot Prompting
Asking an AI to do something without giving it any examples — just a straight instruction. "Write a professional bio for a roofing contractor." This works for simple tasks, but for anything nuanced or brand-specific, few-shot prompting (with examples) usually gets better results.