Site icon Kevin Waites Consulting Services

What ‘we use AI’ means: the three levels of AI agency

When someone tells me “we use AI at our organization,” I’ve learned to ask a follow-up question. It could mean a staff member drafting emails in a chat window. It could also mean agents running workflows across the CRM, the email platform and the shared drive. Those are very different things.

I think about it as a spectrum of agency, meaning how much the AI can actually do on its own. Where your tools sit on that spectrum tells you a lot about where your team’s time is going, and where it’s worth investing next.

The three levels of AI agency

Low agency: the chat window

This is the traditional chatbot or support window. It’s completely reactive and has no context outside the conversation. You give it a prompt and it gives you text. It can’t click a button or touch your data.

Mid agency: copilots inside one app

This is AI built into a single tool, like Canva’s image generator or Notion AI. You give it a prompt and it works inside that tool to get the task done. It’s useful, but it keeps you inside that app’s silo.

High agency: agents that work across your systems

This is where the work itself changes. These agents operate across tools, using connection standards like MCP to reach your CRM, documents and email. They can run a workflow with many steps, and some can write their own code to fill a gap when two tools don’t connect.

The farther along the spectrum you go, the more your role shifts from operating software to directing the work.

Why copilots leave you doing the bridging

Every software company has built an AI copilot into its product by now. Salesforce, Canva and Microsoft all have one, and they’re good at isolated tasks.

The catch is that each one only knows its own app. Your design AI doesn’t know what’s in your CRM, and your email AI doesn’t know what’s in your project management tool. So you end up as the bridge between them. You copy data out of one tool, prompt the AI in another and paste the result into a third.

That’s why I’d call a lot of chatbot prompting digital micromanagement. Leaders who hoped AI would clear their plates find themselves pasting text from a chat window into Word, then into Mailchimp, then fixing the formatting. The chatbot can hand you a strategy, but it can’t log into your CRM and carry it out.

At the high-agency end, you give a goal instead of a string of prompts. “Draft the quarterly donor recap and prepare the list for review” is one instruction, and the agent works through the steps in the background while you grab a coffee.

When the agent writes its own code

There’s one more ability that sets these agents apart from older automation. In a traditional setup, if App A doesn’t have a built-in connection to App B, the workflow breaks. Someone has to step in, hire a developer or find a workaround.

A high-agency agent can write a small script to fetch the data it needs. It does that in a secure sandbox, a separate space away from your live systems, checks that the script works and then finishes the job. Code has become a kind of universal interface for AI, which is a big part of why these agents can adapt to the patchwork of tools most nonprofits run.

How to find where your team sits

  1. List the AI tools your team actually uses. Include the copilots built into software you already pay for, since those often get missed.
  2. Place each one on the spectrum. Is it a chat window, a copilot inside one app, or an agent that works across tools?
  3. Look for the copy and paste. Ask staff where they move text or data between tools by hand. Each of those spots is a place where a person is doing the bridging.
  4. Pick one cross-tool task to move up. The quarterly donor recap is a good candidate, since it touches your CRM, a document and your email tool.
  5. Set limits before you move up. The more an AI can do, the more you need clear rules about what it can change and a person approving anything that goes out.

For leaders: ask what “we use AI” actually means

When a vendor, a board member or your own team says you’re using AI, find out where on the spectrum that use sits. A team with five copilots and no connections between them is still doing most of the coordination by hand.

A few questions to bring to your next team meeting:

Cosly, the AI staff team I’m building for small nonprofits, works at that high-agency end, connecting your separate tools so the work doesn’t stay stuck inside one app.

Exit mobile version