Legal AI Vocabulary Every In-House Leader Should Know

AI-specific terms are being used more widely in legal technology vendor pitches, board updates, procurement reviews, and internal policies. And while a lot of these words sound interchangeable, they aren't. 

An AI agent is not the same thing as an AI chatbot. An AI wrapper is not the same as an AI platform. Generative AI is not in the same category of tool as agentic AI. 

It’s important to understand the difference between terms because they influence what tools you're buying, what they can do for your legal team, and what kind of ROI you’ll be seeing six months into deployment.

So, let’s explore legal AI vocabulary that every in-house leader should know in 2026 and how to differentiate commonly conflated terms.

The Foundations: Core AI Terms

These are the building blocks. What the technology is, before it's wrapped in anything sold to you.

Generative AI 

Generative AI is a category of systems that produce new content based on patterns learned from training data: text, code, images, or audio. In the legal context, generative AI is what powers tools that draft clauses, summarize matters, or produce first-pass responses to recurring legal questions.

Large Language Model (LLM)

Large Language Model (LLM) is the underlying model that powers most generative AI tools currently in use. ChatGPT, Claude, and Gemini are examples of LLMs. In legal applications, the LLM is the engine doing the reading and writing; the product around it determines how it's applied to legal work.

Prompt

A prompt is the instructions given to the model by the user. It could be a question, a drafting request, or a structured set of instructions paired with a document. The specificity of the prompt directly shapes the output.

Context

Context is everything the model can see when generating a response. This includes the prompt, the prior conversation, any attached documents, and any system instructions running in the background. 

Hallucination

A hallucination occurs when an AI model generates fluent and confident output that is factually incorrect. This looks like fabricated citations, misquoted statutes, invented contract terms, etc. It is a structural feature of how LLMs work rather than a defect that gets patched out, which is why human review remains a fixed requirement in any legal application.

The Doers: AI That Takes Action

This is what the AI technology does when it's pointed at actual legal work. 

AI Agent

An AI agent is an AI-powered system that can execute actions, rather than just produce text. An agent can call other software, retrieve data, and complete multi-step tasks. In a legal setting, an agent might intake a request, classify it, retrieve relevant precedent, draft a response, and route it to the appropriate reviewer.

Plug-In

A plug-in is an add-on that extends the capabilities of what a model can do, usually by connecting it to a specific external system. In legal tools, plug-ins often connect AI capabilities to contract repositories or CLMs, matter management systems, e-signature platforms, or communication channels like Microsoft Teams and Slack.

Wrapper

A wrapper is a product built as a thin interface over a third-party model, with limited proprietary logic underneath. Many tools marketed as "AI-powered" in the legal space are wrappers, which becomes relevant during vendor evaluation because the depth of legal-specific functionality varies significantly between wrappers and purpose-built systems.

The Facilitators: The Infrastructure Layer

The layer underneath that determines whether any of ‘the doers’ actually work at scale.

Clean Data

Clean data is structured, accurate, accessible data. AI output quality is bounded by input quality. Most in-house legal data sits in unstructured formats like email threads, PDFs, ad-hoc messages, and spreadsheets maintained inconsistently across the team. Cleaning and structuring that data is often a precondition for AI tools to function as intended.

Guardrails

Guardrails are the technical and policy controls that constrain what an AI system can do, see, and say. They include access controls, output filters, audit logs, and prompt restrictions. For legal teams, guardrails dictate how governance policies become operational inside an AI tool.

AI Legal Front Door

An AI Legal Front Door is foundational infrastructure that allows a legal function to operate in a structured, AI-ready way. It sits at the point where business requests enter legal, capturing the right information up front, triaging by risk and priority, and routing work to the right person, team, or process. Contract management, workflow automation, matter management, analytics, and AI tools across the stack all rely on a functioning Front Door to receive work in a usable form. Without it, the rest of the legal tech stack is operating on incomplete or inconsistent inputs.

Shadow AI

Shadow AI is AI used by employees outside of sanctioned IT or legal channels. Examples include a marketer pasting a draft NDA into a free chatbot, or a sales rep running deal terms through a personal account. Shadow AI is a growing source of confidentiality, privilege, and data security exposure for legal functions.

AI Terms That Get Conflated in Legal Procurement

Three distinctions worth getting right, because they show up in procurement and strategy conversations constantly.

Generative AI vs. Agentic AI

Generative AI produces content, whereas agentic AI takes action. A tool that drafts a clause is generative. A system that receives a request, classifies it, drafts a response, and routes it through a workflow is agentic. Some platforms combine both.

AI Legal Front Door vs. AI Chatbot 

An AI chatbot offers a conversational interface that lets a user ask questions in natural language. A Legal Front Door delivers the structured intake and triage layer that sits behind the conversation, capturing the right information, applying logic, and routing work into downstream systems. In most modern deployments, the chatbot is the surface and the Front Door is the infrastructure underneath. They are integrated, not interchangeable.

Key Takeaways

You don’t need to focus on memorizing definitions. You just need to be able to understand what a legal AI technology vendor is offering and able to deliver. When the pitch says "agent" and the product is a chatbot, you should hear it. When "AI-powered" turns out to mean a wrapper over a publicly available model, you should know what question to ask next.

The legal leaders moving fastest on AI right now aren't necessarily the ones with the deepest technical fluency. Instead, they're the ones who can sit in a vendor meeting, hear the words, and accurately decode what's actually being built. That skill is becoming a procurement filter, a governance filter, and a strategy filter. If you're mapping where AI fits in your legal function, the vocabulary is where the work starts.

Want to learn more? Schedule a call with one of our technology consultants today.