The agents you already own.

Sits beside a user while they write a prompt, suggests refinements, explains why a prompt is producing weak output, and flags prompts that fall outside responsible-AI guidance.

In the sector: the most valuable hour of training you can give a twelve-person nonprofit, and it costs nothing.

Feedback on clarity, coherence, grammar, and tone. Adjusts a document for a named audience, and flags wording that may not travel across cultures or reading levels.

In the sector: grant narratives, donor appeals, and the board memo that has to say a hard thing carefully.

Runs a guided brainstorm through question-based dialogue, builds a session agenda, proposes creative exercises, and helps prioritize what comes out.

In the sector: program design retreats and strategic planning sessions that otherwise consume a facilitator’s fee.

The one to know. Researcher builds a research plan, works recursively across your tenant data and the web for several minutes, and returns a synthesized report with explicit source citations. It can call other agents. Users can choose the model, including Claude, from the model picker. Use it when the output has to be trusted, cited, and shareable, not when you are iterating.

In the sector: funder landscape scans, policy and regulatory reviews, and the competitive analysis a development director would otherwise outsource.

Writes and runs Python against files you give it, up to five at a time, mixing spreadsheets with PDFs and documents. It hypothesizes, tests, and adjusts, and shows its analysis steps so you can review how it reached the answer. No data science staff required.

In the sector: program outcome reporting, grant spend-down analysis, and the year-over-year donor retention question nobody has had time to answer.

Turns a prompt into a structured long-form document, asks clarifying questions before it drafts, and applies your approved Word templates. Files save to OneDrive under your existing policies. It runs only on Anthropic models: if an admin turns Anthropic off as a subprocessor, the agent is hidden, even for licensed users.

In the sector: letters of inquiry, board packets, and policy documents that follow a house structure.

Builds real workbooks, not screenshots of workbooks: working formulas, PivotTables, charts, conditional formatting, and best, base, and worst-case scenario models with the assumptions written out. It runs only on Anthropic models: if an admin turns Anthropic off as a subprocessor, the agent is hidden, even for licensed users.

In the sector: multi-year program budgets, restricted-fund tracking, and the scenario model a CFO builds before every board meeting.

Generates a complete first-draft deck, pulls context from emails, chats, and meeting notes, applies your company template, and turns bullet lists into charts on request. It runs only on Anthropic models: if an admin turns Anthropic off as a subprocessor, the agent is hidden, even for licensed users.

In the sector: the quarterly board update, and the funder site-visit deck assembled the night before.

Joins the meeting as a shared participant, not a private assistant. Builds and tracks the agenda, captures notes and decisions live where everyone can see them, answers questions in the room, and turns discussion into tasks. Requires Loop experiences in Teams to be turned on, and is not supported in meetings with external participants.

In the sector: board and committee meetings, where the decision record is a governance artifact and the secretary is a volunteer.

Real-time speech-to-speech interpretation across up to nine languages, so each participant speaks and listens in the language they choose. Only participants with a Copilot license hear translated audio, and each license includes 20 hours a month.

In the sector: the most underrated agent on this list. Community meetings with limited-English-proficiency participants, and rural hospital care coordination calls where the alternative is a phone interpreter line billed by the minute.

Creates, distributes, and analyzes surveys, turning free-text responses into themes rather than a spreadsheet somebody has to read.

In the sector: constituent feedback, post-program evaluation, and the staff engagement survey that never gets tabulated.

Assists administrators with tenant tasks across Microsoft 365 services. Pre-installed for every Microsoft 365 admin with no added license; what it can do follows the admin’s existing role permissions.

In the sector: useful precisely where IT is one person who also runs the help desk.

Formerly Project Manager. Creates plans, executes tasks, tracks progress, and writes status reports inside Planner.

In the sector: grant-funded project reporting, where the funder wants a milestone update and nobody owns the tracker.

A channel-scoped agent grounded in that team’s conversations and meetings, so the answer comes from what the team actually said.

In the sector: a program team’s institutional memory, which currently lives in one person’s head.

Surfaces expertise across communities so knowledge is findable rather than re-asked.

In the sector: multi-site organizations and hospital systems where the same clinical or operational question is answered independently in four places.

Automates recurring work: sending reminders, managing calendars, and pushing team updates on a schedule.

In the sector: the recurring administrative load that quietly consumes a program coordinator’s week.

Gives leaders and their delegates a current view of the organization for staffing decisions. Requires setup before first use.

In the sector: handle with care. Workforce analytics in a small organization can identify individuals even when the report says it does not.

Recommends skill-building paths against the work someone is actually doing, and runs AI role-play practice. Private preview through the Frontier program; a Viva Learning license is optional and adds content.

In the sector: a plausible substitute for a training budget that does not exist.

A single front door for routine HR and IT requests: checking leave balances, updating personal information, requesting a device. Ships with prebuilt connectors, but it still has to be integrated with the systems that hold the answers.

In the sector: worth it above roughly 150 staff. Below that, the HR generalist is faster than the integration project.