marsten.ai glossary
Don’t let the jargon throw you—we make AI integration simple! But we also recognize that a list of list of phrases and concepts can be super useful to understanding artificial intelligence. For those of us who love to go down rabbit holes, we also recommend checking out these glossaries from The New York Times and MIT.
Last updated: November 18, 2025 (HST)
A
-
A simple experiment that compares two versions of something to see which performs better.
Why it matters
A/B testing helps teams improve results by testing small changes and learning what works best.
Example
You might test two versions of an email subject line to see which one gets more replies.
-
A page section that expands or collapses to show details. Helpful on long pages.
-
A multi-step AI workflow that can use tools like search, email, calendar, or a CRM to help complete a task.
Why it matters
It helps move work forward with less manual copy-and-paste between tools.
Example
An agent might draft a reply, look up a customer, schedule a follow-up, and log a note.
-
A system trained on large amounts of data that can generate text, images, audio, or other outputs based on patterns it has learned.
Why it matters
AI models power tools that can draft content, summarize information, answer questions, and support many business tasks.
-
The approved sources and actions an AI assistant can access, and the sources or actions it is not allowed to use.
Why it matters
Allow-lists and block-lists help reduce risk, protect sensitive information, and prevent unwanted actions or data leaks.
See also
Guardrails, Prompt Hardening
-
A short description of an image used for accessibility and SEO.
Example:
“Douglas Hunter headshot, blue checkered shirt” -
A way for one software system to communicate with another.
Why it matters
APIs let us connect your tools to AI assistants and automations without relying on manual copy-and-paste.
See also
Webhooks, Integration
-
When possible, we prefer Apple-native tools like Shortcuts, Mail, and Calendar before adding new apps.
Why it matters:
simpler stacks, less context switching. -
A set of rules or steps that runs tasks automatically, often triggered by an event.
Why it matters
Automation helps reduce repetitive manual work, improve consistency, and keep routine tasks moving without needing someone to do each step by hand.
Example
When a form is submitted, an automation might draft a thank-you email in your voice.
B
-
Your company’s tone, vocabulary, and style. Helpers are tuned to match it.
See also:
voice and style pack, style guide, few-shot example
C
-
A simple list of improvements to each helper so changes are visible and reversible.
-
Breaking long documents into smaller pieces so an AI assistant can read, search, and retrieve information more accurately.
Why it matters
Smaller chunks make it easier for the system to find the most relevant information and stay within the model’s context limits.
See also
Context Window
-
Following laws and company policies for privacy and security.
Why it matters:
keeps builds safe and aligned with your counsel and IT. -
Plain-English lines in proposals or emails that explain how we will use specific data, tools, or testimonials.
-
The amount of text an AI model can consider at one time when generating a response.
Why it matters
A larger context window allows the AI to keep more background information in view, which can improve accuracy and continuity.
-
A custom, role-specific AI assistant from Marsten AI that works inside your existing tools to help offload repetitive tasks, surface clear insights in your voice, and operate with strong privacy guardrails.
-
Settings that control how steady or varied the writing is.
Tip:
lower is steadier, higher is more exploratory. -
The next step you want someone to take.
Example:
“Book your 15-minute discovery call.” -
A tailored AI assistant configured for your business using your instructions, examples, approved tools, and workflow needs.
Why it matters
A custom helper is more useful than a generic one because it can reflect your process, your standards, and the way your team actually works.
See also
Helper, Voice & Style Pack
D
-
How an organization manages data access, quality, retention, and compliance across its systems.
Why it matters
Strong data governance helps protect sensitive information, improve data quality, and ensure systems use information responsibly.
See also
Data Inventory, Retention, Least Privilege
-
A list or catalog of the data sources an organization stores or uses.
Why it matters
A clear data inventory helps teams understand what information exists, where it lives, and how it should be protected or used.
Examples
Customer records, internal documents, CRM data, support tickets, or financial reports.
-
Using only the minimum amount of data needed to complete a task.
Why it matters
Limiting data reduces risk, protects sensitive information, and supports stronger privacy practices.
See also
De-Identification, Redaction
-
Where data is stored geographically. Some companies require specific regions.
-
Removing or altering personal details so data cannot easily be linked back to a specific person.
Why it matters
De-identification helps protect privacy and reduces the risk of exposing sensitive personal information.
See also
PII, PHI, Redaction
-
A short, complimentary conversation to confirm fit and identify the biggest opportunity.
-
A contract add-on that sets privacy, retention, and subprocessors for a project.
E
-
A numeric representation of text that helps an AI system understand meaning and find related content.
Why it matters
Embeddings help AI assistants search for similar ideas quickly, even when the wording is not exactly the same.
See also
Vector Database, RAG
-
Protecting data by encoding it both while it moves across networks and while it is stored on servers or disks.
Why it matters
Encryption helps prevent unauthorized access and keeps sensitive information secure during transmission and storage.
-
A small set of test examples used to evaluate an AI assistant’s accuracy, tone, and reliability before launch.
Why it matters
Eval sets help confirm that the assistant behaves as expected before it is used in real workflows.
See also
Human-in-the-Loop, Guardrails
F
-
FWhen a helper cannot proceed, it hands off to a person or a simpler workflow so users are not stuck.
-
A small number of examples included in a prompt to show the AI the format, style, or tone you want.
Why it matters
Providing examples helps the AI produce outputs that are more consistent with your expectations.
Example
Two sample replies in your voice, followed by “write one for this case.”
-
Further training an existing AI model on a small set of your examples to improve style, accuracy, or performance for a specific use case.
Why it matters
Fine-tuning can help a model better match your domain, tone, or workflow.
-
Your unique way of speaking and deciding. Helpers are tuned to this “signal” so outputs sound like you.
G
-
Google’s analytics tool for measuring site traffic and conversions.
See also:
tracking link (UTM) -
An answer that cites approved sources so you can verify it.
See also:
grounding, RAG -
Keeping an AI assistant’s answers tied to approved sources and trusted business information.
Why it matters
Grounding helps improve accuracy and reduces the chance of the AI making things up.
See also
Grounded Answer, Knowledge Base
-
Rules and limits that control what an AI assistant can read, write, and do.
Why it matters
Guardrails help keep AI useful, safe, and under human control, especially when working with business information.
Example
Guardrails can limit access to approved documents and require human review before any email is sent.
H
-
When an AI model sounds confident but gives an incorrect or made-up answer.
Why it matters
Hallucinations can create confusion, errors, or risk, especially in client-facing or business-critical work.
How we reduce it
Grounding, retrieval, clear examples, and human review for anything client-facing.
-
A small, purpose-built AI assistant configured for a repeatable task in your voice.
Why it matters
Helpers can speed up routine work, reduce back-and-forth, and cut down on unnecessary edits.
-
A process where a person reviews, edits, or approves AI output before it is used or sent.
Why it matters
Human oversight helps maintain accuracy, quality, and appropriate tone, especially for client-facing communication.
Default
Human review is the default for client-facing messages.
I
-
The standing guidance that sets tone, role, and boundaries for a helper.
-
Connecting tools so they can share data, trigger actions, and work together more smoothly.
Why it matters
Integrations reduce manual handoffs, improve consistency, and help workflows move faster across the systems your team already uses.
See also
API, Webhooks
J
-
Attempts to make an AI model ignore its instructions, break its rules, or reveal information it should not share.
Why it matters
These attacks can lead to unsafe behavior, inaccurate output, or exposure of sensitive data.
See also
Prompt Hardening, Guardrails
K
-
A collection of approved documents, notes, and reference materials that an AI assistant can use to answer questions and support its work.
Why it matters
A strong knowledge base helps AI give more accurate, relevant, and consistent answers based on trusted business information.
See also
Retrieval, RAG
L
-
The delay between sending a request to a system and receiving the response.
Why it matters
Lower latency makes AI assistants feel faster and more responsive during everyday use.
-
A security principle that gives people or systems only the minimum access needed to complete their work.
Why it matters
Limiting access reduces the risk of accidental changes, data exposure, or misuse of sensitive information.
See also
RBAC, Access Approvals
-
A type of AI model trained on large amounts of text that can generate language by predicting likely next words.
Why it matters
LLMs power tools that can draft content, answer questions, summarize information, and support many business tasks.
-
Keeping a record of operations to help debug and audit. You can opt out of nonessential logs.
M
-
An extra sign-in step, like a code, that protects accounts even if a password is stolen.
N
-
A confidentiality agreement to protect shared information. Often used before deeper data access.
O
-
The steps to get set up: access, data samples, voice examples, and a quick success plan.
-
Running AI in a private environment so prompts and outputs are not retained by a public vendor.
See also:
private mode, zero-retention mode -
A simple standard operating procedure that shows when to use the AI assistant, the steps to follow, and what good output looks like.
Why it matters
A one-pager helps teams use the assistant consistently, reduce confusion, and get better results with less back-and-forth.
-
Any file or result a workflow produces, like a document, email draft, or spreadsheet.
P
-
Health-related information that can be linked to a specific person and is protected under healthcare privacy laws.
Why it matters
PHI must be handled carefully to protect patient privacy and comply with healthcare regulations.
Examples
Medical records, diagnoses, treatment information, insurance details, or any health data connected to a person’s identity.
-
Information that can be used to identify a specific person.
Why it matters
Protecting PII helps safeguard privacy and reduces the risk of exposing sensitive personal data.
Examples
Names, phone numbers, email addresses, government ID numbers, or home addresses.
-
A limited trial with real users to validate a workflow before broader rollout.
See also:
rollout, eval set -
Using settings or deployments where prompts and outputs are not stored by vendors or are stored only in your environment.
See also:
zero-retention mode, on-prem -
The instruction or input you give an AI model to guide what it produces.
Why it matters
Clear prompts lead to more accurate, useful, and consistent results.
Example
Instead of asking “write something about our company,” a better prompt might be “draft a short follow-up email thanking a client for today’s meeting and summarizing the next steps.”
-
A focused review session to improve prompts, identify quick wins, and make AI outputs more useful and reliable.
Why it matters
A prompt audit helps uncover where small changes can improve clarity, consistency, and overall results.
Includes
Before-and-after examples to show what changed and why.
-
Techniques used to make prompts more resistant to prompt injection, misuse, or unintended data exposure.
Why it matters
Prompt hardening helps protect the system, reduce risky behavior, and keep sensitive information from being exposed.
Examples
Allow-lists, strict tool permissions, and content filters.
-
A reusable prompt with placeholders that can be filled in for different situations.
Why it matters
Prompt templates help teams get consistent results and save time by reusing prompts that already work well.
Example
“Write a friendly two-paragraph email in our voice to [name] about [topic].”
Q
-
A useful improvement you are not doing yet because no one connected the dots. Often a small helper with a big payoff.
R
-
A method that lets an AI assistant read approved documents or trusted business information before answering.
Why it matters
RAG helps keep answers grounded in your actual content, which improves accuracy and reduces guessing.
See also
Grounding, Vector Database
-
Caps vendors set on how many requests you can make per minute. We design around these.
-
Access permissions that are granted based on a person’s or system’s role.
Why it matters
RBAC helps keep permissions organized and secure by ensuring each person or AI assistant only has access to the information they need.
Example
A finance role might have read-only access to billing records, while other roles cannot view them.
See also
Least Privilege
-
The process an AI system uses to work through a task and decide what to output.
Why it matters
Good reasoning helps improve accuracy and reliability, especially on more complex tasks.
Note
We focus on strong results rather than exposing every internal step.
-
The process of masking or removing sensitive details before text is shared, stored, or used by an AI system.
Why it matters
Redaction helps protect private or confidential information and reduces the risk of exposing sensitive data.
Example
A phone number might be replaced with [REDACTED] before a note is saved or shared.
-
How long data is kept before it is deleted.
Why it matters
Clear retention rules help reduce unnecessary storage, protect sensitive information, and support better privacy practices.
Default
We prefer short retention timelines and deletion at project end unless you require otherwise.
-
The job or function an AI assistant is asked to take on when helping with a task.
Why it matters
A clear role helps the assistant respond in the right tone, use the right context, and stay focused on the job it is meant to do.
Example
“You are a helpful client services assistant.”
-
The process of introducing a new AI assistant or workflow to more users in stages after the pilot.
Why it matters
A staged rollout helps reduce risk, gather feedback, and make sure the new workflow works well before broader adoption.
Example
A team might start with one department, review the results, and then expand the workflow to the rest of the organization.
-
A step-by-step plan for introducing AI assistants into a team or workflow.
Why it matters
A structured rollout helps ensure the assistants are launched safely, the team understands how to use them, and results can be measured and improved over time.
Example
A rollout playbook might include pilot testing, team training, usage guidelines, and simple metrics to track impact.
S
-
Rules that prevent harmful or disallowed outputs. Applied at the vendor level and in our prompts.
-
The precise boundaries of what a helper will and will not do. We keep scopes small and useful.
-
Short-term context a helper can remember during a conversation to stay consistent, then reset.
See also:
context window -
Third-party security standards some vendors have. We prefer tools with these attestations when relevant.
(System and Organization Controls 2 / International Organization for Standardization 27001)
-
Plain-English steps so a task is done the same way every time. Often delivered as a one-pager.
-
A short document listing scope, deliverables, timeline, and responsibilities.
-
The tiniest build that delivers value now. We favor this over big-bang projects.
-
Sending output as it is generated so you see it appear quickly.
-
A brief reference for tone, formatting, and dos and don’ts so outputs stay on-brand.
See also:
brand voice, voice and style pack -
The official place a piece of information lives, like your CRM for client notes.
See also:
single source of truth
T
-
How long it takes from kickoff to your first measurable result. We track and work to shorten it.
-
A small chunk of text models count to measure length and cost. Roughly three to four characters on average.
-
Letting a helper call an external tool mid-task, like search, calendar, or email.
-
A link with tags such as utm_source and utm_medium so we can see which campaign brought a visitor.
See also:
GA4, QR code -
What a model learned from before you ever used it. For IP-sensitive work, we choose tools with clear training sources.
-
A trigger is the event that starts a workflow. An action is what happens next.
V
-
A specialized database that stores embeddings so an AI system can quickly find content that is meaningfully similar to a question or request.
Why it matters
Vector databases help AI assistants search large sets of documents more accurately, which improves retrieval and supports better answers.
See also
Embedding, RAG
-
Labeling iterations of prompts, workflows, and models so we know what is running where.
-
A compact guide with examples, preferred phrasing, and clear dos and don’ts that helps AI assistants stay on-brand.
Why it matters
A strong voice and style pack helps AI produce content that sounds more like your business and requires fewer edits
W
-
An automatic message one tool sends to another when something happens, such as a form submission or a new customer record.
Why it matters
Webhooks help systems respond in real time so workflows can move faster without manual handoffs.
See also
API, Integration
Z
-
A vendor setting where prompts and outputs are not stored or used for training.
See also:
private mode, on-prem -
How many examples you give in the prompt. Zero-shot is none, one-shot is one example, few-shot is a small set.