AI agents
AI Agents for Nonprofits and Growing Businesses: What to Build, What to Buy, and Where
Operations used to follow the software. Now organizations can design the work first, with people and AI agents on the same team.
Published by 2N2 Solutions ·6 min read
The renewal notice arrives with a price increase and a three-year term. The tool does its job. It also asks the team to work its way: the intake form has the fields the product wanted, the follow-up runs on the product's schedule, and the report shows what the product decided to count.
Then someone on the operations team mentions, almost in passing, that she built a small agent last month. It drafts the follow-up after every new inquiry, in the organization's own voice, and puts it in front of a person to send. It took her two afternoons.
That moment is happening in a lot of organizations right now. It raises a question that used to have an obvious answer: what should we buy, and what should we build ourselves?
Short answer: buy the work everyone does the same way. Configure the platforms you already run. Build where the work is how you serve people and how you grow. The organizations getting this right are redesigning how their work runs, with people and AI agents on the same team, and choosing the tools after.
Should we build or buy AI agents?
For most of the last two decades, building meant developers, long timelines, and budgets most growing businesses and nonprofits didn't have. So operations followed the product. You bought the system, and the team learned to work the way it worked.
That order is reversing. In Pax8's Q3 2026 SMB AI Pulse Report, released October 12, 2026, 67% of AI-engaged businesses said they had built one or more AI solutions now in regular use. The survey covered 1,150 leaders at companies of 5 to 499 employees across the United States, the United Kingdom, Germany and Sweden. Pax8 sells cloud software to small businesses through IT providers, so read it as a vendor's survey; it is still one of the few current looks at organizations this size.
The same report shows the other side. Among U.S. builders, 36% have formal, documented review of what they built, and another 41% review consistently without writing it down. The building is outpacing the operating model around it. That is the real decision in front of leaders, and it is bigger than build or buy.
Buy, configure, or build: a guide by kind of work
- Buy where the work is the same for every organization: payroll, email, accounting, the core of your CRM. The market has solved these, and your edge is not in doing them differently.
- Configure where the platform you already pay for can carry the work with the right setup: the agents, flows and permissions your existing suite offers. Most organizations use a fraction of what they already own.
- Build where the work defines how you serve and how you grow: how a lapsed donor is welcomed back, how a client moves from the sale to the first delivery, what leaders see before a funding decision.
- Build small and in place. A working agent inside the systems you run, reviewed by a person, beats a new platform that asks the team to start over.
- Decide who runs it before you build it. Every agent needs a named person who reviews what it produces and keeps it working as the business changes.
Copilot Studio, Power Automate or Claude: what building looks like on each stack
The stack matters less than people expect. What matters is building where your team already works. Here is what that looks like on the two stacks we see most.
Microsoft 365. For organizations already running Microsoft 365, Copilot Studio is where agents get built. Microsoft describes it as a way to build agents and "connect them to your organization's data and systems," with prebuilt or custom connectors and tools that take action (Microsoft Learn). Power Automate handles the flows that move work between apps and services (Microsoft Learn). Dataverse holds business data in tables, with role-based security controlling who can see what (Microsoft Learn). The question people search most, Copilot Studio or Power Automate, usually has a plain answer: Power Automate for a defined sequence of steps, Copilot Studio when the work needs judgment, language, or a conversation, and often both together.
Model choice now sits inside that stack too. Microsoft enables Claude models from Anthropic by default for most commercial-cloud customers, outside the EU, EFTA and UK, and admins control whether they are available (Microsoft Learn). An organization can choose the model workflow by workflow without leaving its environment.
Claude. For teams building outside a single suite, Claude Code is an agentic tool that "reads your codebase, edits files, runs commands," available in the terminal, IDE, desktop app and browser (Claude Code docs). The Model Context Protocol (MCP) is "an open-source standard for connecting AI applications to external systems," which is how an agent reaches the accounting system, the shared drive, or the CRM (modelcontextprotocol.io). This is the route when the work spans systems from different vendors, or when the organization wants its tools to stay portable as models change.
The CRM you already run. Whatever holds your donor, member or customer records is usually the right home for agents that touch those relationships. Build there, inside its permissions, rather than copying records somewhere new.
How do you build AI agents for a nonprofit?
The same way any organization should: start with the work, not the tool. For a nonprofit, that usually means the work closest to the mission and the money. Two places come up again and again: the second conversation with a first-time donor, and the funder report someone assembles by hand each quarter. Pick the one your team most wants back, build it where the team already works, and keep a person reviewing what goes out.
Reskilling the team for people-and-agent work
Every stack asks the same new skills of the people using it: giving an agent clear direction, reviewing its work with a skeptical eye, and writing down what was learned so the next run starts from a better place. They are management skills, applied to a new kind of teammate. The organizations moving fastest teach them to every team, not only to the person who built the first agent.
Filling the gap
Most organizations sit in a gap right now. The tools can do more than the operation is set up to use. The team is willing, and busy. The answer is rarely another product. It is builders who work inside the operation and the platforms it runs on, and who build with the team, so every solution is one the team helped shape, from the first conversation to a system running in production.
Which part of your work would you design differently if the tools were no longer the limit? If that question is on your 2027 list, let's talk it through.
Sources
- Pax8, "Pax8 Research Finds Small Businesses Are Building AI Faster Than They Can Govern It," Q3 2026 SMB AI Pulse Report, released October 12, 2026. Survey of 1,150 small business leaders at companies of 5 to 499 employees in the United States, United Kingdom, Germany and Sweden; commissioned by Pax8, a cloud software distributor. Read the release. Accessed 2026-11-11.
- Microsoft Learn, "Overview: Microsoft Copilot Studio." Read on Microsoft Learn. Accessed 2026-11-11.
- Microsoft Learn, "Explore the Power Automate home page." Read on Microsoft Learn. Accessed 2026-11-11.
- Microsoft Learn, "What is Microsoft Dataverse?" Read on Microsoft Learn. Accessed 2026-11-11.
- Microsoft Learn, "Claude models from Anthropic in Microsoft Online Services." Read on Microsoft Learn. Accessed 2026-11-11.
- Anthropic, "Claude Code overview." Read the documentation. Accessed 2026-11-11.
- Model Context Protocol, "What is the Model Context Protocol (MCP)?" Read on modelcontextprotocol.io. Accessed 2026-11-11.