60 terms you will hear when a vendor pitches AI, a board asks about it, or your team wants to build with it. Each one is defined in plain English first, then explained the way an owner needs it: why it matters to the business, what it looks like in practice, and the mistakes that cost companies time and money.
What a model is, what it costs to run, and where it can live.
The techniques that turn a general model into something that knows your business.
Software that takes actions, not just answers, and how to keep it accountable.
The plumbing: where your data lives and how systems talk to each other.
What can go wrong, which rules apply, and the controls that hold up.
The decisions an owner or executive actually signs off on.
How AI changes how buyers find you and how your ad spend learns.
An agentic workflow is a business process in which an AI model handles some of the steps itself, deciding how to complete them, while the overall sequence, rules, and handoffs are defined in advance. It sits between rigid automation, which follows fixed rules, and a fully autonomous agent that plans everything on its own.
An AI agent is software that uses a language model to pursue a goal by choosing its own next steps: it reads the situation, picks a tool or action, checks the result, and repeats until the task is done or it needs a person. Unlike a chatbot, it acts inside your systems instead of only answering.
An AI chatbot is a conversational interface, usually on a website, app, or messaging channel, that uses a language model to understand questions written in everyday language and reply in kind. Modern chatbots can answer from a company's own content and hand off to people, but on their own they converse rather than take actions in business systems.
An AI coding agent is an AI system that performs software engineering tasks with some autonomy: reading a codebase, writing and changing code, running tests, fixing failures, and opening changes for review. Unlike autocomplete assistants that suggest the next line, a coding agent can carry a task from a ticket to a proposed pull request. Claude Code is one example.
Evals are structured tests that measure how well an AI system performs on a defined task, using a set of real or realistic inputs with known good outcomes. They are scored by rules, by another model acting as a judge, or by people, and they are rerun whenever prompts, models, or data change to catch regressions before users do.
AI governance is the set of policies, roles, processes, and controls a company uses to decide which AI systems it builds or buys, how they are approved, how risks are assessed, and how they are monitored once running. It assigns accountability for AI decisions and makes sure use complies with law, contracts, and the company's own standards.
Guardrails are the controls around an AI system that keep its inputs, outputs, and actions within acceptable limits: input filtering, output checks for policy, format, and sensitive data, limits on which tools and data it can reach, spending caps, and rules that route risky cases to a person. Good guardrails are enforced in software, not merely requested in prompts.
A hallucination is output from an AI model that is fluent and confident but false or unsupported: an invented fact, a made-up citation, a policy that does not exist, or a number that appears nowhere in the source. It happens because language models generate plausible text rather than looking facts up, so it can be reduced but not fully eliminated.
An AI knowledge base is a company's documents, policies, procedures, and records organized so an AI assistant can search them and answer questions with citations. It combines the content itself, kept current by clear owners, with a retrieval system that respects permissions. Staff or customers ask in plain language instead of hunting through folders and wikis.
AI orchestration is the coordination layer that decides which model, tool, data source, or agent handles each step of a task, in what order, and what happens when a step fails. It covers routing, passing context between steps, retries, approvals, and logging, so that many AI components behave like one dependable system.
An AI readiness assessment is a structured review of whether a company is prepared to get value from AI and where to start. It examines business processes and pain points, data quality and access, systems and integrations, security and compliance constraints, team skills, and leadership alignment, then produces a prioritized list of opportunities with estimated value, cost, and risk.
Answer engine optimization (AEO) is the practice of structuring a website's content so AI assistants and answer features, such as ChatGPT, Perplexity, Google's AI Overviews, and voice assistants, can find it, understand it, and use it in direct answers. It emphasizes clear question-and-answer content, concise definitions, structured data, and pages machines can read cleanly.
API integration is connecting software systems through their application programming interfaces (APIs), the defined ways one program can request data from or send instructions to another. It lets your CRM, ERP, billing, scheduling, and AI systems share data and trigger actions automatically, instead of people re-keying information between screens or exporting spreadsheets.
Build vs buy is the decision between developing custom software, including AI systems, and purchasing or subscribing to an existing product. Buying is usually faster and cheaper for standard needs. Building makes sense when the process is a competitive advantage, when off-the-shelf tools do not fit, or when long-term cost, control, or data ownership favor owning it.
A business associate agreement (BAA) is a contract required by HIPAA between a covered entity, such as a provider or health plan, and a vendor that creates, receives, maintains, or transmits protected health information on its behalf. It obligates the vendor to safeguard the PHI, limit its use, report breaches, and pass the same obligations to its subcontractors.
A Chief AI Officer (CAIO) is the executive accountable for a company's AI strategy, adoption, and governance: deciding where AI should be applied, prioritizing and funding use cases, setting policy and risk controls, coordinating data and technology teams, and measuring results. The role can be full time, combined with another executive role, or filled on a fractional basis.
Claude is a family of large language models made by Anthropic, available through the Claude apps, an API, and major cloud platforms including Amazon Bedrock and Google Cloud Vertex AI. The family comes in tiers (Opus, Sonnet, and Haiku) that trade capability against speed and cost. Businesses use Claude for writing, analysis, coding, and agents.
Claude Code is Anthropic's agentic coding tool. It works in a developer's terminal, editor, or browser, reads an existing codebase, plans changes, edits files, runs commands and tests, and can prepare commits, asking permission for actions according to its settings. It connects to other tools through the Model Context Protocol and can be scripted into automated workflows.
Computer vision is the field of AI that lets software interpret images and video: detecting objects, reading text, measuring, classifying, and spotting anomalies. It ranges from specialized models trained for one task, such as finding defects on a production line, to general multimodal models that can describe and reason about almost any image.
A context window is the maximum amount of text, measured in tokens, that a language model can take into account in a single request, including your instructions, any documents or conversation history, and the answer it writes. Anything outside the window is invisible to the model for that request, as if it never existed.
A conversions API is a server-to-server connection that sends conversion events, such as leads, purchases, or qualified opportunities, from your own systems directly to an advertising platform, instead of relying only on a browser pixel. Meta's Conversions API is the best-known example; Google, TikTok, and LinkedIn offer comparable server-side or offline conversion options.
A customer relationship management (CRM) system is the software a company uses to track its relationships with customers and prospects: contacts, companies, deals, communications, support history, and pipeline stages. Salesforce and HubSpot are widely used examples, and some companies build custom CRMs. It is usually the system of record for revenue activity.
A data warehouse is a central database designed for analysis and reporting, where data from many operational systems, such as ERP, CRM, billing, and marketing platforms, is collected, cleaned, and organized so it can be queried together. Snowflake, Google BigQuery, Amazon Redshift, and Databricks are common platforms. It is the usual foundation for dashboards and data-driven AI.
EHR integration is connecting an electronic health record system, such as Epic, Oracle Health, or athenahealth, with other software so patient, scheduling, clinical, and billing data can flow between them. It typically uses healthcare standards like HL7 version 2 messages and FHIR APIs, along with vendor-specific interfaces, under HIPAA's privacy and security requirements.
An embedding is a list of numbers that represents the meaning of a piece of text, an image, or other data, produced by an embedding model. Items with similar meaning get similar numbers, so software can find related content by measuring the distance between embeddings, even when the wording is completely different. Embeddings power semantic search and recommendations.
ETL stands for extract, transform, load: the process of pulling data out of source systems, cleaning and reshaping it into a consistent format, and loading it into a destination such as a data warehouse. A common modern variant, ELT, loads raw data first and transforms it inside the warehouse. Both keep reporting and AI fed with reliable data.
Fine-tuning is the process of further training an existing AI model on a curated set of your own examples so it consistently produces a particular style, format, or behavior. It changes the model's weights, unlike prompting or retrieval, which only change what the model is given at the moment it answers. It suits narrow, repeated tasks.
A fractional CTO is an experienced technology executive who serves a company part time or through a firm, owning the decisions a full-time chief technology officer would own: technical strategy, architecture, build versus buy, vendor and hiring choices, budget, and delivery oversight. Companies use one when they need senior technical judgment without a full-time executive hire.
Generative AI is the category of artificial intelligence that creates new content, such as text, images, audio, video, or code, in response to a prompt, rather than only classifying or predicting from existing data. Large language models and image generation models are the most widely used examples, and most current business AI tools are built on them.
Generative engine optimization (GEO) is the practice of improving how often and how accurately a brand, product, or source appears in answers produced by generative AI systems such as ChatGPT, Claude, Gemini, Perplexity, and Google's AI features. Academic researchers introduced the term in 2023. It overlaps heavily with answer engine optimization and builds on traditional SEO.
HIPAA and AI refers to how the US Health Insurance Portability and Accountability Act applies when AI systems create, receive, store, or transmit protected health information. There is no special AI exemption: covered entities and their business associates must meet the same Privacy, Security, and Breach Notification Rule requirements, including a business associate agreement with any AI vendor handling PHI.
Human in the loop is a design pattern in which a person reviews, approves, or corrects an AI system's output at defined points before it takes effect. The human might approve every action, only exceptions above a risk threshold, or a sample for quality control. It is how companies get AI speed without handing over final accountability.
Inference cost is what it costs to run an AI model to produce outputs, as opposed to the cost of training it. With hosted models it is billed per token of input and output; with self-hosted models it is the compute, hardware, and operations needed to serve requests. For most companies it is the main recurring AI expense.
Internal tools are software applications a company builds or configures for its own staff rather than its customers: admin panels, approval workflows, operations dashboards, quoting tools, inventory trackers, and custom CRMs. They replace spreadsheets, email chains, and manual steps with purpose-built screens connected to company data, and increasingly include AI features.
A large language model (LLM) is an AI model trained on very large amounts of text to predict the next piece of text, which lets it write, summarize, translate, classify, extract information, and reason through problems in everyday language. Claude, GPT, Gemini, and Llama are families of LLMs, the engine inside most modern AI products.
Lead scoring is ranking prospective customers by how likely they are to buy or how valuable they would be, so sales and marketing focus on the best opportunities first. Rule-based scoring assigns points for attributes and actions, such as company size or a pricing page visit; predictive scoring uses a model trained on which past leads actually converted.
The Model Context Protocol (MCP) is an open standard, introduced by Anthropic in 2024, for connecting AI applications to external tools and data. An MCP server exposes capabilities such as tools, resources, and prompts in a standard format, so any compatible AI client can use them without a custom integration for each pairing.
A multi-agent system is a setup in which several AI agents, each with its own role, instructions, and tools, work together on a task. Typically an orchestrator agent breaks the work into parts and hands them to specialist agents, then combines or checks their results. It trades simplicity for parallel work and separation of duties.
Multimodal AI is AI that can take in or produce more than one kind of data, such as text, images, audio, video, or documents, within the same model or system. A multimodal model can read a photo of a damaged part, a scanned invoice, or a chart and reason about it together with written instructions.
On-premise AI means running AI models and the systems around them on hardware in your own facilities or private data center, rather than calling a vendor's cloud service. Data is processed entirely inside infrastructure you own. It is typically chosen for strict data residency, air-gapped environments, regulatory requirements, or very high, steady workloads.
An open-weight model is an AI model whose trained parameters, called weights, are published so anyone can download and run it on their own hardware or cloud, subject to its license. Llama, Mistral, Qwen, DeepSeek, and Gemma are examples. Open weight is not always open source: training data and code are often withheld, and licenses vary.
A private LLM is a language model deployment in which your data and prompts stay inside an environment you control or have contractually isolated, rather than a shared consumer service. It can mean an open-weight model you host yourself, or a commercial model accessed through your own cloud account with no training on your data and defined retention.
Prompt engineering is the practice of writing and refining the instructions, context, and examples given to an AI model so it produces reliable, useful output for a specific task. It covers stating the goal and audience, supplying the right reference material, specifying the output format, giving examples, and testing changes against real cases rather than guessing.
Prompt injection is an attack in which instructions hidden in content an AI system processes, such as a web page, email, document, or user message, trick the model into ignoring its original instructions. It can make an AI leak data, take unintended actions, or produce harmful output. OWASP lists it first among security risks for LLM applications.
A proof of concept (PoC) is a quick, limited build that shows an idea can work, usually on sample data without real users. A production system is one the business depends on: it runs on live data, handles errors and edge cases, and is secured, monitored, and maintained. The gap between the two is where most AI projects stall.
Protected health information (PHI) is individually identifiable health information held or transmitted by a HIPAA covered entity or its business associate, in any form. It covers information about a person's health, care, or payment for care that is linked to identifiers such as a name, address, dates, phone number, email, medical record number, or photo.
A reasoning model is a language model trained to work through a problem step by step before giving its final answer, spending extra computation on intermediate thinking. This improves results on multi-step tasks such as math, analysis, planning, and code, at the cost of slower responses and more tokens per answer. Many providers let you adjust how much it thinks.
Retrieval-augmented generation (RAG) is a technique in which an AI system first searches your own documents or data for passages relevant to a question, then gives those passages to a language model to write its answer. The model answers from your current, approved information instead of relying only on what it learned in training, and can cite sources.
The ROI of AI is the measurable business return from an AI initiative relative to its full cost. Returns come from labor hours saved, faster cycle times, fewer errors, higher conversion or retention, or new revenue; costs include building or licensing, integration, usage fees, change management, and upkeep. It is measured against a baseline taken before launch.
Robotic process automation (RPA) uses software bots that mimic a person's clicks and keystrokes to follow fixed rules across applications. AI automation uses models that interpret unstructured inputs like emails, documents, and images and make judgment calls. RPA excels at stable, repetitive screen work; AI automation handles variation that rules cannot anticipate.
Shadow AI is the use of AI tools by employees without the knowledge, approval, or oversight of the company's IT, security, or compliance teams, such as pasting customer data into a personal chatbot account or connecting an unvetted AI plugin to work email. It is usually well intentioned and almost always a sign of unmet demand.
SOC 2 is an attestation report, defined by the American Institute of Certified Public Accountants (AICPA), in which an independent CPA firm examines a service organization's controls against the Trust Services Criteria: security, availability, processing integrity, confidentiality, and privacy. A Type I report covers control design at a point in time; Type II covers operating effectiveness over a period.
A system prompt is the standing set of instructions an application gives a language model before any user input, defining its role, rules, tone, available tools, and limits. Users usually never see it. It shapes every response in a product, which makes it one of the most important parts of an AI feature to write, version, and test.
Technical debt is the future cost created when software is built in a faster or easier way instead of a sounder one: shortcuts, missing tests, outdated dependencies, tangled code, and undocumented decisions. Like financial debt, it charges interest, making every later change slower and riskier until it is paid down. The metaphor comes from programmer Ward Cunningham.
A token is the unit of text a language model reads and writes: a whole word, part of a word, a number, or a punctuation mark. Model limits and provider pricing are both measured in tokens. A common rule of thumb for English is that one token averages about three quarters of a word, though it varies by model.
Tool use, also called function calling, is the ability of a language model to request that your software run a specific function, such as looking up an order, querying a database, or sending a message, with structured inputs the model fills in. Your code runs the function and returns the result, and the model continues from there.
Total cost of ownership (TCO) is the full cost of a technology decision over its useful life, not just the purchase price or build quote. For software and AI it includes licenses or development, implementation, integration, data preparation, training, hosting and usage fees, maintenance, support, security and compliance work, and the eventual cost of switching away.
A vector database is a database designed to store embeddings and quickly find the ones most similar to a query, a task called similarity or nearest-neighbor search. It is the usual storage layer behind semantic search and retrieval-augmented generation. Dedicated products exist, and many standard databases, including PostgreSQL with the pgvector extension, now offer the same capability.
Vibe coding is building software by describing what you want to an AI tool in plain language and accepting the code it generates, largely without reading or understanding that code. The term was coined by AI researcher Andrej Karpathy in early 2025. It is fast for prototypes and personal tools, and risky for anything handling customers, money, or sensitive data.
Workflow automation is the use of software to carry out the steps of a business process, such as moving data between systems, routing approvals, sending notifications, and updating records, without a person doing each step by hand. Traditional automation follows fixed rules; AI workflow automation adds models that read documents, classify requests, and handle variation.
Every definition leads with a short answer you can repeat in a meeting. The full page for each term goes further: the business case, a worked example, and what to watch for when a vendor or an internal team proposes it. Where a term touches work we do, the page links to the matching service so you can see how it is scoped and priced.
Looking for more than definitions? The resources hub collects the blog and the AI workflow playbooks, and direct answers on cost and timelines live on the FAQ.
tell us what keeps you up at night.
Scoped by the people who ship it. Priced before we start.
book a call drop your number