# 41 terms a buyer meets, each with a sourced example

Glossary, read 8 Nov 2026

**The short answer.** Forty terms from the agent conversation: agent, assistant, workflow step, approval step, audit log, off switch, sensitive field, MCP server, tool, OAuth, read-only, write action, confirmation, scope, Agent System of Record, ISO 42001, high-risk system and the rest, each with an example read from a vendor's own page on 8 Nov 2026.

- [Acts in other systems](/glossary/acts-in-other-systems/): The matrix's column for an agent that does something outside the HRIS once approved: sends an email or a Slack message, fires a webhook, exports to payroll, provisions an account.
- [Agent audit log](/glossary/agent-audit-log/): A log that records which agent acted, what it did, on which record, when and who approved it, exportable by actor and date.
- [Agent builder](/glossary/agent-builder/): A tool for a non-engineer to describe an agent in plain language and have it exist: the process, the trigger, the approver.
- [Agent Capability Matrix](/glossary/agent-capability-matrix/): The 17-by-12 table at the center of this site: 17 platforms read against 12 capabilities on their own pages, each cell yes, partial or none found with the vendor's sentence and the page it was read from on 8 Nov 2026, plus an MCP status column.
- [Agent Diary](/glossary/agent-diary/): The site's step-by-step reading of three processes (onboarding a hiring wave, a compensation cycle, employee questions from policy) through every platform whose matrix cell is yes or partial: the trigger, the draft, the approval, the actions in other systems, the log entry and the off switch, each quoting the vendor's page.
- [Agent Gateway](/glossary/agent-gateway/): Workday's entry point through which outside agents (Claude, Gemini, OpenAI, Copilot) reach its Agent-Ready Tools over MCP.
- [Agent System of Record](/glossary/agent-system-of-record/): Workday's name for a register in which every agent (Workday's, a partner's or a customer's) is enrolled, given an identity and permissions, governed and audited, with Agent Analytics on what it did.
- [Agents under control](/glossary/agents-under-control/): The criterion weighted 30 on this site: whether an agent runs a People process end to end and whether a person approves before it goes live, sees what it did in a log, can switch it off and can keep it from sensitive fields.
- [AI governance](/glossary/ai-governance/): The matrix's twelfth column: whether the vendor publishes an AI management certificate (ISO 42001 held or in progress), an AI trust center, or a named framework for the AI it ships.
- [Answers questions from policy](/glossary/answers-questions-from-policy/): The matrix's first column: an assistant that answers an employee's question from the handbook, the policies or the record.
- [Approval step](/glossary/approval-step/): The point at which an agent stops and a person says yes before a record is written, a message leaves, money or access changes, or a process starts for a group.
- [Assistant](/glossary/assistant/): A chat or sidebar that answers a question from policy, the help center or the record and stops there.
- [Automated employment decision tool](/glossary/automated-employment-decision-tool/): New York City Local Law 144's term for a computational process that issues a simplified output used to substantially assist or replace discretionary decisions in hiring or promotion.
- [Bias audit](/glossary/bias-audit/): An independent third party's evaluation of an automated employment decision tool's selection rates by sex and race or ethnicity, required annually by New York City's Local Law 144 before the tool is used, with a summary published.
- [Confirmation on write](/glossary/confirmation-on-write/): A server rule that returns every write to the user for a yes before anything changes, which is an approval step on the connector side.
- [Connector](/glossary/connector/): The client-side entry for an MCP server: the item a user adds in Claude's connectors or ChatGPT's apps that points at the vendor's server and signs in.
- [High-risk AI system](/glossary/high-risk-ai-system/): Under Regulation (EU) 2024/1689, Article 6(2), an AI system listed in Annex III, which for employment covers recruitment and selection, decisions on promotion and termination, task allocation based on behavior or traits, and monitoring and evaluation of performance.
- [HR agent](/glossary/hr-agent/): Software that takes a People process from a trigger to an approval without a person building each step, acts in other systems once approved, logs what it did, and can be switched off.
- [Human in the loop](/glossary/human-in-the-loop/): A rule, in a product or in a contract, that a person reviews or decides before an AI output takes effect.
- [ISO 42001](/glossary/iso-42001/): ISO/IEC 42001:2023, the management system standard for artificial intelligence: policies, roles, risk assessments, controls and reviews for the AI an organization builds or uses, audited by a third party.
- [MCP server](/glossary/mcp-server/): A server a vendor publishes so that an outside AI client (Claude, ChatGPT, Gemini, Cursor) can call the vendor's tools under the connecting user's permissions through the Model Context Protocol.
- [MCP status](/glossary/mcp-status/): One of five values on the matrix, read from the vendor's own page only: live for all customers, beta or early access, announced, third-party wrapper only, none found.
- [None found](/glossary/none-found/): The matrix's value when a capability was looked for on the vendor's pages on the read date and not found.
- [OAuth 2.0](/glossary/oauth-2-0/): The authorization standard most MCP servers on the matrix use: the client gets a token for the user after the user signs in to the vendor, so the client never holds a password and the token carries the user's permissions.
- [Off switch](/glossary/off-switch/): A setting that turns an AI feature off for the company, a role or a feature.
- [Openness](/glossary/openness/): The criterion weighted 25 on this site: how far outside AI tools reach the platform's data, scored on an official MCP server and its status, read or write access, OAuth, a public API, webhooks and documented connectors.
- [Partial](/glossary/partial/): The matrix's value when a vendor's page documents a narrower form of a capability: part of a process, a control for some tools, a log that does not name the agent, a switch that scopes rather than turns off.
- [Pilot](/glossary/pilot/): A bounded trial of one agent on one process for one team with the four controls on, ending with one number: the People hours not spent.
- [Read date](/glossary/read-date/): The date a vendor's page was read for a cell, a status or a score, printed in the eyebrow of every page and beside every source.
- [Read-only access](/glossary/read-only-access/): An MCP server whose tools return data and never change it.
- [Runs a comp cycle](/glossary/runs-a-comp-cycle/): The matrix's test for an agent that runs a compensation cycle: proposes the lines, routes them to managers, stops at approval and writes the result.
- [Runs onboarding](/glossary/runs-onboarding/): The matrix's test for an agent that takes a hiring wave from the hire event to the approved record, the tasks, the documents and the policy assignments without a person building each step.
- [Scope](/glossary/scope/): The set of data categories or permissions an MCP server or an agent is allowed to reach, set by an admin.
- [Sensitive-field control](/glossary/sensitive-field-control/): A rule that hides fields such as salary, medical notes or ratings from a role, a view or an agent even when the rest of the record is visible.
- [Service desk agent](/glossary/service-desk-agent/): An AI agent that sits in Slack, Teams or a portal, answers employee questions, routes tickets and runs plugins in other systems, without holding the employee record itself.
- [Tool](/glossary/tool/): A named function an MCP server exposes to a client, with inputs, outputs and a permission: list employees, get time off balance, submit a review.
- [Trigger](/glossary/trigger/): The event that starts an agent or a flow: a hire event from the ATS, a date, a request, a change to an employee field.
- [Vendor claim](/glossary/vendor-claim/): A statement on a vendor's own page about speed, scale or outcome that this site could not check against a second source; it is quoted with the label and caps the criterion it supports at 7.
- [Vibe View](/glossary/vibe-view/): Shapes's name for a dashboard, report, tool or workflow built from a plain-language description on live, permission-aware data.
- [Workflow with an AI step](/glossary/workflow-with-an-ai-step/): A flow a person built in which one step is drafted by AI: a plan, a document, a policy, a report.
- [Write action](/glossary/write-action/): An MCP tool that changes the record: creates a request, updates a field, submits a review, runs a workflow.

Last reviewed 10 Nov 2026. Canonical: https://bestaiagentsforhr.com/glossary/
