Words we use when we talk about AI at work.

An access boundary is the edge around a part of company knowledge that decides which roles, people, skills, and agents may read or change it.

An agent is a digital teammate with a defined role, clear responsibilities, its own identity, and strict limits on what it can see and do.

Automation is a defined action or workflow that runs from a clear trigger and known rules, usually without a person doing each step by hand.

A company context library is the shared, access-controlled body of knowledge that explains how a business works, makes decisions, handles exceptions, and judges good work.

Durable context is the company's approved, lasting knowledge: the pages people and AI should treat as authoritative until a reviewed change replaces them.

Earned autonomy is the process of allowing a specific kind of agent action to run automatically only after the record shows it has been reviewed, tested, and approved reliably.

An eval is an automated evaluation that checks whether an AI skill or agent is producing work that meets the standard before and after it is allowed to run.

A learning is a lesson worth keeping, captured from the work so the company can review it and decide whether it should become official context.

Memory is the searchable index built from company knowledge so people, skills, and agents can find the right context quickly while the readable source files remain authoritative.

A Protobase Agent is an individual employed agent inside a company's ProtobaseOS instance, with a named role, written responsibilities, tests, permissions, and human oversight.

ProtobaseOS is the AI operating system Protobase builds and runs around a company's own knowledge, workflows, permissions, skills, agents, and memory.

A skill is a small, well-defined AI workflow that turns approved company context into a specific kind of work, such as drafting a proposal section or preparing a lead review.