AI, tech and product glossary for product managers
31 concepts PMs run into every week, explained without needless jargon: a precise definition, why it matters to you, key points, common mistakes and the courses that go further.
A
- A/B testProduct and data
An A/B test is a controlled experiment in which users are randomly assigned to two or more versions of a product experience, and a metric chosen in advance is compared between the groups.
- Acceptance criteriaProduct and data
Acceptance criteria are the conditions a piece of work has to satisfy before users and the business can call it finished.
- AI agentAI
An AI agent is a system in which a language model decides, step by step, which actions to take to reach a goal: it calls tools, reads the results, then chooses the next action until it judges the task done or hits a limit.
Evals are systematic tests of an AI feature's outputs against defined criteria, run on a set of representative inputs.
AI guardrails are the controls placed around a language model to keep a product's behavior within acceptable limits.
- APITech
An API (application programming interface) is a defined way for one piece of software to request data or actions from another.
Authentication verifies who a user or a system is: a password, a passkey, a one-time code, or delegating the check to a provider such as Google.
C
The context window is the maximum amount of text, counted in tokens, that a language model can take into account in a single request: instructions, conversation history, documents and its own answer included.
D
- Data modelProduct and data
A data model describes the information a product stores and how it fits together: the entities (users, accounts, orders), their attributes, and the relationships between them, such as one account having many users.
E
An embedding is a list of numbers, a vector, that represents the meaning of a piece of text or an image, so that items with similar meanings get vectors close to each other.
I
- IdempotencyTech
An operation is idempotent when performing it several times has the same effect as performing it once.
L
LLM as a judge is an evaluation method in which a language model scores or compares the outputs of an AI feature against a rubric.
An LLM hallucination is an output that sounds plausible but is false, unsupported or invented: a citation that does not exist, a wrong figure, a feature the company never shipped.
M
Introduced by Anthropic in November 2024, MCP (Model Context Protocol) is an open specification that defines how AI applications connect to external tools and data sources.
A merge conflict happens when Git cannot combine two branches automatically because both changed the same part of a file, or one branch removed a file that the other edited.
N
Non-functional requirements describe how well a system must work rather than what it does: performance, availability, security, scalability, accessibility, maintainability.
- North Star metricProduct and data
A North Star metric is one number a product team adopts to express the value its users receive, in a way that also predicts the company's long-term success.
O
- Outcome-based roadmapProduct and data
An outcome-based roadmap organizes product work around the results the team wants to achieve, such as more customers activating in their first week, instead of a feature list with ship dates.
P
- Product analyticsProduct and data
Product analytics is the collection and analysis of data about how people use a product: which actions they take, in what order, how often and whether they come back.
Prompt engineering is the practice of writing and refining the instructions given to a large language model so that it produces the output you need, reliably.
Prompt injection is an attack in which text supplied to a language model contains instructions that override or divert the ones set by the developer.
- Pull requestTech
A pull request, called a merge request on GitLab, is a proposal to merge the changes from one branch into another, usually the main branch.
R
Retrieval-augmented generation (RAG) is an architecture where an application first searches a document collection for the excerpts that best match a question, then sends those passages to a large language model along with the question.
Role-based access control (RBAC) is an authorization model in which permissions are attached to roles, such as admin, editor or viewer, and users receive one or more roles.
S
Software architecture is the set of structural decisions about a system: which components exist, what each one is responsible for, how they communicate and where data lives.
- SQL for product managersProduct and data
SQL (Structured Query Language) is the standard language for querying relational databases and data warehouses.
- Statistical significanceProduct and data
A result is statistically significant when the observed difference would be unlikely if the change had no real effect, given a threshold chosen in advance.
Structured outputs are model responses that follow a format defined in advance, typically JSON matching a schema, instead of free text.
T
- Technical debtTech
Technical debt is the future cost created when a team chooses a faster or simpler solution now over a better one that would take longer.
Tool calling, also called function calling, lets a language model request an action from the application around it.
W
- WebhookTech
A webhook is an HTTP request that a service sends automatically to a URL you provide when an event happens, such as a payment succeeding or a file being uploaded.