AI Agents for Nonprofits: the Three Real Products and Where They Fit

An AI agent is software that carries out a multi-step task on its own, within rules you set, rather than waiting to be asked each time. For nonprofits, three real products exist today: Salesforce Agentforce Nonprofit, Blackbaud’s Development Agent for Raiser’s Edge NXT, and Bonterra Que. Each only works inside its own vendor’s system, none publishes a price apart from Salesforce’s ten free licences, and adoption is still tiny: in the Nonprofit Tech for Good survey of 826 nonprofits, 1.2% were using agentic AI for fundraising. This page explains what counts as an agent, compares the three products, sets out which nonprofit tasks suit an agent and which do not, and gives a sequence for trying one without putting donor relationships at risk.

What counts as an agent, and what does not

“Agent” is now applied to almost every AI feature, so it helps to place tools on a ladder. The difference that matters is not intelligence but who acts.

Most of what nonprofits use today sits on the first two rungs. A chat assistant like Claude or ChatGPT does nothing until you type. Agents start at the fourth rung, where the software plans steps and carries them out, and become autonomous at the fifth, where it sends messages and replies without approving each one. The further down the table, the more of your organization’s judgement and voice you are delegating.

The three nonprofit agent products

Salesforce Agentforce Nonprofit. Announced on 2 December 2025 as the new name for Nonprofit Cloud, with four prebuilt agents. The Prospect Research agent prepares a briefing on a high-value donor’s past engagement inside Slack; Participant Management summarises a client’s background and services for a case manager and can create goals, take notes and flag referrals; Volunteer Capacity and Coverage finds unfilled shifts and matches volunteers by skills, availability and location; Donor Support answers common donor questions. Eligible nonprofits get ten free licences through the Power of Us programme. The case studies in the announcement describe expected time savings rather than measured ones.

Blackbaud Development Agent. Part of Blackbaud’s “Agents for Good”, built into Raiser’s Edge NXT. Blackbaud describes it as a proactive virtual team member that engages donors at scale with your oversight, through a guided sequence of messages from introduction to engagement or solicitation. You configure its brand voice, the knowledge it draws on, its display name and data mapping; it schedules messages, escalates conversations to staff, and logs its communication history. This is the clearest example of a fifth-rung agent sold to nonprofits: it writes to donors on your behalf, with staff stepping in on escalated conversations.

Bonterra Que. Launched on 1 October 2025 across Bonterra’s products, including EveryAction and Network for Good. It is organised as skills rather than named agents: building segments from a plain-English request, drafting forms and emails that a person approves, surfacing stewardship insights, and on the grants side matching and drafting proposals. Most of what it does sits on the fourth rung.

The common thread is lock-in. Each agent works inside its own vendor’s database and nowhere else, so for most organizations the choice of agent has already been made by the choice of CRM.

How many nonprofits use agents

Very few. The Nonprofit Tech for Good survey, completed by 826 nonprofits and published in January 2026, found 1.2% using agentic AI for fundraising, against 4.5% using smart donation forms and 2.3% using predictive AI to identify likely mid-level or major donors. Surveys of general AI use put adoption far higher, mostly individual staff using chat assistants, which is covered in nonprofit AI statistics.

The organizations already running AI in production report the practical problems. In Coastal’s 2026 survey of 75 nonprofits with AI live, 72% struggled with data accuracy or availability once it was running, 56% faced more maintenance than planned, and 53% said staff did not trust the outputs enough to act on them. An agent amplifies whatever is in the database it works from. If addresses are stale and gift histories incomplete, it will act on stale and incomplete information, faster.

What an agent is good for, and what it is not

The pattern: agents suit tasks that are high volume, rule bound and cheap to correct. They suit relationships poorly, because a nonprofit’s relationships with major donors and with the people it serves rest on judgement and trust, and an agent’s errors are sent under your name.

Chatbots and voice agents are agents pointed at the public. A website chatbot that answers “how do I update my monthly gift” or “where is my receipt” is a reasonable use, provided it hands over to a person on anything else and never invents a policy. A voice agent calling donors is a different matter: check state telemarketing and charitable solicitation rules before any automated call, and disclose that the caller is not a person.

Agents and recommenders for fundraising teams

For most small development teams, the useful step is not an autonomous agent but the third rung: software that reviews the whole donor file on a schedule and tells a person who needs attention and why, then drafts the message for a person to edit and send. It delivers most of an agent’s time saving while keeping the relationship in human hands.

Gratefully, an AI donor intelligence and stewardship platform built for nonprofits, is built on that design. It reads Salesforce for Nonprofits, Raiser’s Edge NXT, Bloomerang, Little Green Light or HubSpot, with documents and email engagement, works through the portfolio each night, and produces a ranked list each morning with the reason for each donor. Its Plays, such as a thank-you after a first gift or a check-in with lapsing donors, are on by default: its assistant, Grace, drafts the emails overnight into a Ready to Send queue, a person reviews and approves each one, and it goes out from that person’s own Gmail or Outlook with their signature. That is the fourth rung of the ladder above, stopping deliberately short of the fifth; donor figures come from a query layer rather than the language model. It is free for one person and $79 a month billed annually for Essential. It is wrong for an organization that wants outreach sent without a person approving each message, which it deliberately does not do; for Raiser’s Edge NXT users who want that kind of autonomy, which Blackbaud’s own agent offers; for files with little history to read; and for anyone needing an issued SOC 2 Type II report, which it expects in the fourth quarter of 2026.

If your CRM is Salesforce, Raiser’s Edge or Bonterra, look at the vendor’s own agent first, because it already sits on your data. If it is something else, the honest position in 2026 is that no general purpose agent can safely run donor outreach on your behalf, and a recommender plus a person is the better design.

How to try an agent without risking donor relationships

1. Clean the data first. Duplicates, deceased and moved records, missing consent flags. Every agent problem in the Coastal survey starts here.
2. Start read only. Briefings and summaries before anything that sends. You learn how often it is wrong without anyone outside seeing it.
3. Pick one bounded task. Volunteer shifts, receipt questions, or a mid-level sequence to a test segment, not “fundraising”.
4. Keep approval on until the error rate is known. Review the first hundred actions. Record each correction.
5. Write the rules into policy. Which tasks may run unattended, what must escalate, who is accountable for what the agent sends. The AI use policy template has a tool register for exactly this.
6. Tell people. Donors and volunteers should know when they are corresponding with software.
7. Measure outcomes, not hours. Time saved is easy to claim. Retention, response and complaint rates tell you whether it worked.

Staff who have done a short course in working with AI make far better supervisors of an agent; free options are compared in AI training for nonprofits.

How this page was put together

Product details were read on 9 October 2026 at Salesforce’s Agentforce Nonprofit announcement, Blackbaud’s Development Agent help centre and Bonterra’s launch release. Availability dates are as each vendor stated them; we could not confirm whether every planned agent has since shipped, and say so where relevant. Adoption figures are from the Nonprofit Tech for Good survey (826 nonprofits, January 2026) and Coastal’s 2026 AI operations survey (75 nonprofits). No vendor publishes agent pricing apart from Salesforce’s free licences. Gratefully is a company we cover closely; its details are from its own site.

This is reference information, not legal or tax advice. Rules vary by state and change over time. For a decision that carries real consequences, check the current text at irs.gov or your state registry, and talk to a nonprofit attorney or CPA.