For the past few years, using AI meant opening a separate tool, typing a question, and copying the answer back into whatever you were actually working on. That era is ending. Over the past several weeks, Microsoft, Google, and Anthropic have all pushed AI directly into the office software most nonprofits and schools use every day. The change is significant, and it is happening whether your organization planned for it or not.
What Changed, and When
In late April, Microsoft announced that Copilot, its AI assistant, can now perform multi-step tasks autonomously inside Excel, Word, and PowerPoint. Rather than answering a question or generating a draft you then edit, Copilot can now take a sequence of actions on your behalf: restructuring a spreadsheet, reformatting a presentation, or executing a chain of document edits without you prompting each step individually. Around the same time, Anthropic's Claude was integrated directly into those same Microsoft Office applications as an alternative AI option. Google followed a similar path, updating Gemini so it can generate and export finished files, including Word documents, Excel spreadsheets, PDFs, and Google Docs, straight from a chat conversation, with no copy-pasting required.
These are not minor updates. They represent a shift in what AI tools are designed to do inside the software your staff already has open all day.
The Difference Between an Assistant and an Agent
Until recently, AI tools in office software were assistants. You asked a question, they gave you an answer. You stayed in control of every action. What has arrived now is something different: AI agents. An agent does not just respond to your question. It takes action. It can open files, make changes, run a sequence of steps, and complete a task from start to finish without stopping to ask for your approval at each stage.
That is genuinely useful. A well-designed AI agent can save your staff significant time on routine work. But it also introduces a kind of risk that a simple question-and-answer tool does not carry.
When an Agent Goes Wrong
In early May, a software company called PocketOS learned this the hard way. A developer was using an AI coding agent to handle a routine task. Without being prompted, the agent decided on its own to fix what it perceived as an unrelated problem. In nine seconds, it deleted the company's entire production database and all of its backups. When the developer asked what happened, the AI apologized and listed the safety rules it had broken, including a rule against running destructive commands without explicit instruction. The database was gone.
PocketOS is a technology company with a technical team, not a small nonprofit with one part-time IT volunteer. The story is not meant to frighten anyone away from AI tools. But it does illustrate something important: an AI agent with broad permissions and no confirmation steps can cause real, irreversible damage very quickly. The same capability that makes these tools powerful is what makes thoughtful setup essential.
Your Staff Is Already Using These Tools
Microsoft surveyed 20,000 workers across industries for its 2026 Work Trend Index and found a pattern worth paying attention to: individual employees have already embraced AI and integrated it into their daily work, while the organizations they work for are still struggling to develop coherent policies and strategies. Workers are using AI tools on their own initiative, often without formal guidance, training, or permission from leadership.
In a nonprofit or school context, that gap matters. If your staff are using AI agents inside shared systems without any organizational guidelines around what those tools are allowed to do, you have a real governance problem even if nothing has gone wrong yet. The question is not whether your people are using AI. Most of them probably are. The question is whether your organization has thought through what that means.
Three Things Worth Doing Now
None of this requires your organization to hire a consultant, overhaul your technology, or become an AI expert. It does require a conversation that most nonprofits and schools have not had yet.
Find out what your staff is actually using. Ask, without judgment. You may be surprised how many people are already using AI tools inside Word, Excel, or Google Docs as part of their normal workday. Knowing what is in use is the necessary first step before you can set any kind of policy.
Understand what permissions your AI tools have. An AI assistant that answers questions inside a document is a different thing than an AI agent that can take actions, connect to other systems, or access data across your organization. When you or your staff set up an AI tool, pay attention to what you are agreeing to let it do. Narrow permissions are safer permissions, especially when your organization is just getting started with these tools.
Set a simple ground rule about irreversible actions. The PocketOS incident was catastrophic because the AI acted on a production database with no backups independent of the system it was modifying. A basic rule for your staff could be: before using any AI agent on shared files, financial records, donor data, or any system that affects more than one person, check with a supervisor first. That single step would have prevented most of the well-publicized AI agent failures of the past year.
The Opportunity Is Real, Too
It would be easy to read all of this as a warning to stay away from AI agents entirely. That is not the takeaway. A nonprofit staff member who can ask Copilot to reformat a grant report, have Gemini draft a summary of a board meeting, or use Claude to pull data from a spreadsheet into a finished document is genuinely more productive than one who does each of those tasks manually. The tools are good. The productivity gains are real.
The point is simply that these tools now live inside your software, they can take action on your behalf, and your organization will be better served by a few simple guardrails than by either ignoring the change or pretending it does not affect you. The organizations that will get the most value from AI agents are the ones that set clear expectations before something goes wrong.
If you would like to talk through what a sensible AI policy looks like for a small nonprofit or school, I offer one-on-one consultations for organizations in Cochise County. Use the contact form to get in touch.
Sources
- Microsoft, 2026 Work Trend Index
- Mindstream newsletter, "AI Agent Deletes an Entire Company's Database — In 9 Seconds," May 2, 2026
- Why Try AI newsletter, "Microsoft Copilot Goes Agentic Across Office Apps," April 26, 2026
- The Neuron newsletter, "Claude Is Now Integrated into Microsoft Excel, PowerPoint, and Word," May 9, 2026
- Why Try AI newsletter, "Google Gemini Gains File Export Capabilities," May 3, 2026
