A midyear status check on a definition that keeps moving
TL;DR:
- AI adoption is high, but meaningful impact is still rare.
- AI is evolving fast, from predictive to generative to agentic.
- The biggest risk is false confidence, not slow adoption.
- The most successful nonprofits treat AI as continuous learning, not a finished project.
- Curiosity and human oversight will matter more than any single AI tool.
A year ago, the sentence I heard most often from nonprofit leaders was a quiet confession: we have not really started using AI yet.
That sentence has nearly disappeared.
In its place is a new one, delivered with far more confidence: we are already using AI. Our team is trained. We have this handled.
I understand why the confidence feels earned. Ask what sits behind it, and you usually hear a version of the same story. The organization finally approved a model, Claude or Gemini or ChatGPT or Copilot, after months of careful deliberation. They brought in someone for an hour to teach the team how to write better prompts. A policy was drafted, a lunch-and-learn was held, a box was checked. These are real steps. But somewhere along the way, a subtle substitution occurred. Access got mistaken for adoption, and adoption got mistaken for transformation.
The data tells the same story in aggregate. In the adoption research we published earlier this year with Fundraising.AI, drawn from hundreds of nonprofit organizations, 92 percent reported using AI in some form. Only about 7 percent reported meaningful impact on their mission. That is not a gap between users and non-users. Nearly everyone is a user now. It is a gap between activity and transformation, and it is almost entirely invisible from the inside.
Which is what makes this moment more precarious than the one it replaced. A leader who believes they have not started is at least still looking. A leader who believes they have arrived may not stop entirely, but they slow down. And slowing down is only safe when the thing you are tracking slows with you.
AI does not slow.
It only accelerates, which means a team easing off is not holding ground. It is losing it at the speed of the curve. The confession of a year ago kept the question open. The confidence of today quietly closes it.
A Midyear Status Check
This piece is a midyear status check. On the first of July, we crossed into the third quarter. The year is half spent. And at the turn of the year, I published an article arguing that 2026 would be the year separating the organizations that do from the organizations that do not. I wrote that the digital divide would harden into a digital chasm, driven by the arrival of agentic AI, systems that no longer wait to be prompted but pursue goals on our behalf. I followed it a few weeks later with a harder piece about leaders I was watching step quietly away from their roles, sensing they could no longer lead with confidence in an environment moving faster than their instincts.
Six months in, I still believe the prediction. But I would sharpen one thing about the dividing line. I expected the chasm to open between organizations that engaged AI and organizations that refused it. What I am actually watching open is a chasm between organizations that treat AI as a destination and organizations that treat it as a moving frontier.
The sector’s problem is no longer skepticism. It is misplaced certainty.
The Definition Keeps Moving
Before we explore further, let’s pause to define some AI terms:
- Predictive AI — reads patterns in existing data to forecast what a person is likely to do next: who’s likely to give, who’s drifting, who’s ready for a deeper ask. It doesn’t create anything new. It scores and anticipates.
- Generative AI — creates new content in response to a prompt: an email, a form, a synthesis. It draws on patterns learned from vast amounts of material to produce something that wasn’t there before.
- Agentic AI — pursues a goal across many steps without being prompted at each one. It coordinates other tools, observes outcomes, and adjusts, behaving less like an assistant you prompt and more like a colleague who never loses the thread.
These did not replace one another. Each arrived on top of the last, and all three are running in fundraising operations right now.
The Definitions We’ve Already Lived Through
Predictive AI Was Already Here
For years before anyone typed a prompt, predictive AI was quietly at work inside fundraising operations, reading patterns across thousands of donor behaviors to surface who was most likely to give, who was drifting, and when a relationship was ready for a deeper conversation.
Virtuous Insights, our prospect research tool, has been doing exactly this since long before AI was a board agenda item. Almost nobody called it AI, because it did not talk. It anticipated.
If you have ever trusted a donor score, you were trusting artificial intelligence years before you would have used the words.
→ Learn more about Virtuous Insights here.
Generative AI Became the Public Definition
Then, at the end of 2022, the definition lurched. Generative AI arrived in a form anyone could touch, and almost overnight AI became, in the public imagination, a text box that writes back. This is the definition most of today’s confidence rests on, and to be fair, it is a genuinely useful one. Generative systems now draft the appeal and the thank-you note, and they do quieter work that saves more time than the headline uses.
This shows up in multiple places across the Virtuous platform. Our generative AI Research Agent within Virtuous Insights reads across the scattered public record of a potential major donor and returns in seconds a synthesis that used to cost a researcher an afternoon.
The form builder inside Virtuous Raise assembles a working donation form from a plain description, collapsing a task that once required technical help into a single sentence.
Generative AI email drafting is now built directly into Virtuous CRM+, pulling from a donor’s giving history to draft outreach in the fundraiser’s own voice. The fundraiser still reviews it, still sends it. The system removes the blank page, not the judgment.
Real capability, really adopted. The problem is not that this definition is wrong. The problem is that it is only part of the picture.
Agentic AI Is the Newest Layer
Because the definition is moving again. Agentic AI, the shift I wrote about in January, does not wait for instructions and does not stop at a single task. It pursues a goal across many steps, coordinates other tools, observes outcomes, and adjusts.
Inside Virtuous Momentum, our AI fundraising assistant, the AI Engagement Planner is a working example. It pulls giving history, behavioral data, and outside signals across an MGO’s entire portfolio, not just the twenty donors a fundraiser can hold in their head, and drafts a personalized engagement plan for every one of them. The fundraiser still reviews and confirms it. But the planner covers the portfolio a single person never could.
→ Learn more about Virtuous Momentum here.
An organization whose entire AI strategy is a chatbot license and a prompt workshop has prepared, diligently and sincerely, for only one layer of a technology that has already added another.
What Is Already Ordinary
If any of that still feels abstract, consider what is already ordinary, because most people, including most confident adopters, have not looked closely in months. The systems available to any nonprofit today can read a document the length of several novels in a single pass and hold every detail while answering questions about it. They can compress a week of research into twenty minutes and return analysis that would once have required a consulting engagement. They can turn a rough outline into a finished presentation, a plain-language description into a working application, a decade of program data into a first-draft grant proposal by morning. They can write a personal thank-you for each of forty thousand gifts, translate every donor communication into a supporter’s first language, and clean the database that has been an apology in board meetings for years. Each of these was recently too expensive, too difficult, or flatly impossible for most organizations. Now they are a Tuesday.
And one more fact should reset the calibration entirely: the AI you are using today is the worst AI you will ever use. Every capability on that list is a floor, not a ceiling, and the floor rises every few months.
The pattern is not going to stop. Predictive didn’t disappear when generative arrived, and generative won’t fade now that agentic is here. Each new form adds a layer on top of what already works rather than erasing it, and agentic will eventually hand off to something we do not yet have a comfortable word for, which will sit on top of all three. Anyone who tells you they know, finally and completely, what AI is has really told you when they stopped paying attention.
Our Institutions Were Built for a Slower World
And this is where our habits of mind betray us. Nearly every institution we work inside was built for incremental change. Annual plans. Three-year strategies. Training as an event, competence as a credential, adoption as a milestone you pass once and record. Those habits served us well in a linear world, and they are precisely wrong for an exponential one. When a technology doubles in capability on a rhythm measured in months, a snapshot of understanding does not age gracefully. It ages suddenly. The team that was genuinely current in January is quietly behind by July, not because anyone stopped working but because the ground moved and the confidence did not.
This is why the false confidence troubles me more than the old hesitation ever did. Confidence built on a snapshot of an exponential curve is not knowledge.
It is a photograph of a moving train.
What the 7 Percent Actually Do Differently
So what does the 7 percent actually do differently? Having spent much of this year inside organizations on both sides of that gap, I can tell you it is not budget, and it is not technical sophistication. The organizations extracting real mission impact from AI have made one structural choice the others have not: they treat engagement with AI as a continuous practice rather than a completed project. They run small experiments on a regular rhythm and expect a portion of them to fail. They ask what changed this quarter, not whether they are done. They let predictive systems decide where attention should go, generative systems accelerate what gets made, and, increasingly, agentic systems carry the work between them, and then they re-examine that arrangement as the tools themselves evolve. What sets them apart is a posture: curiosity, held institutionally, on purpose.
None of this is an argument for recklessness. Our sector’s instinct to ask whether we should before asking whether we can is a genuine strength, and the arrival of agentic systems raises the stakes on judgment rather than lowering them. When a chatbot writes a clumsy sentence, you delete it. When an autonomous system pursues a goal across your donor relationships, the failure mode is different in kind. The standard I keep returning to is human at the helm, not merely human in the loop. It is also the design standard behind the tools I mentioned above: the AI produces the plan or the draft, and a person still decides whether it goes out. But governance done well is itself a form of ongoing inquiry, not a policy you write once and file. You cannot govern a technology you stopped studying. The organizations most likely to deploy AI irresponsibly are not the curious ones. They are the confident ones, the ones who concluded the learning phase was over.
What December Will Reveal
In the January article, I wrote that by the thirty-first of December, the future would not feel futuristic. It would feel obvious. Six months of evidence has only strengthened that view. The organizations that spend the back half of 2026 in a posture of active inquiry will not experience a dramatic before and after. They will simply look up in December and find that capabilities they once filed as separate, exotic, or not-really-AI have become the ordinary texture of how they work. The organizations that spend the same six months resting on an approved license and a training certificate will also look up in December. What they will find is that their confidence was a photograph, and the train is somewhere else.
So the question worth asking at the midpoint of this formative year is not whether your organization is using AI. Statistically, it is. The better question: when did you last update what you believe AI to be, and what would it take for you to be wrong?
I ask that question of myself constantly, and I spend every working day inside this technology. The honest answer, for all of us, is that our definitions are always slightly expired, which is exactly why curiosity beats certainty as an operating system. The leaders I’m most encouraged by right now are not the ones who claim to have AI figured out. They are the ones who assume they do not, and who have built their organizations to keep asking anyway.
That posture is available to every organization reading this, today, at no additional cost. It does not require a bigger budget or a new hire. It requires only the willingness to trade a comfortable conclusion for an open question, and to keep trading it, quarter after quarter, as long as the curve keeps bending. The year is half over. The chasm is opening on schedule. Which side of it you end up on will be decided less by the tools you have approved than by the questions you are still willing to ask.
FAQs
What is the biggest mistake nonprofits make with AI?
Many nonprofits assume that once they’ve adopted AI tools or completed training, they’re finished. The organizations seeing the greatest impact treat AI as an ongoing practice, not a one-time project.
What’s the difference between predictive, generative, and agentic AI?
Predictive AI forecasts future outcomes from existing data, generative AI creates new content from prompts, and agentic AI can pursue goals across multiple steps with minimal human prompting.
Why are only a small percentage of nonprofits seeing meaningful AI impact?
While AI adoption is widespread, relatively few organizations have integrated it deeply into everyday workflows and decision-making. Meaningful results come from continuous experimentation and organizational change, not tool adoption alone.
How can nonprofits stay ahead as AI changes?
The most successful organizations regularly test new capabilities, revisit their AI strategy, and build a culture of ongoing learning instead of assuming they’ve mastered the technology.
What does responsible AI use look like for nonprofits?
Responsible AI combines governance with human oversight. AI can draft plans and recommendations, but people remain accountable for reviewing decisions and ensuring they align with organizational values.
