I winced as the third presentation that morning had the exact same purple to blue gradient as the first two. Same rounded-corner icons and the same slide titled “Key Objectives” with a verb, a noun, and a phrase about needing “community engagement”. I was in the back of the room at an in-person convening in late August trying to follow the substance and realizing there wasn’t much to follow. Literally some of the presenters did not seem to know what was on their slides.

Separately, I’d observed a team had submitted a proposal that described end-to-end their plan and how they would run the logistics of an event: venue management, transport coordination, participant communications, vendor contracting. On paper it looked super thorough and used all the right terms. At the time, I suspected it was AI generated but even if it was, I thought that they had matched the pitch against what they could actually do. I hoped it was a beautiful presentation of their vision but that it was still their own vision of the event.

They were given the go ahead to run the event and what they could actually do, day to day, with the staff and systems they had, was much thinner. 4 weeks before the event was supposed to happen, I was stepping into threads I wasn’t supposed to be in and calls that were supposed to have been led by them. I’ve seen plenty of events that didn’t go perfectly before, but what felt different this time was the distance between the proposal and the team who’d supposedly written it.

Figuring out what to call that gap is what I’ve been sitting with these last few days.

The difference between the picture perfect (generic) powerpoint presentations and what the presenters actually knew and could say. The difference between what the vision and execution plan for the event versus reality.

This isn’t about these people specifically (they worked hard with what they had and I appreciate what they tried to do!). In fact, as someone who works with a tiny team of 2.5 people managing a network of hundreds, I deeply empathize with the need to use technologies to bolster capacities and manage all the work that needs to get done.

The bigger part I’m grappling with is something I haven’t heard named in the dominant critiques of AI. I still don’t have a name for it, I’d love to hear if you do.

It’s around what AI now does for organizations trying to sound fundable in 2026 (in a world where the world of philanthropy is literally shifting under us). A small team can generate a proposal that reads as if it has a medium operations unit. A volunteer network can produce a deck that looks like it came from a communications department. The tools are free, fast, and fluent in the language funders expect and at least initially the veneer can hold.

But the point where human judgment has to enter the loop doesn’t disappear just because an AI tool is used to write the proposal or answer emails. So at 6am when the van doesn’t show up, someone still has to call the driver and decide what to tell thirty people waiting in a lobby. Someone needs to have held the overall vision in their head and be able to convene and gather an event organizing team around it and delegate, keep track of timelines and details, make sure all the small small things are thought of. If that person doesn’t have the capacity, time, or standing inside their own organization to say “this isn’t working”, the capacity and knowledge gap becomes visible eventually.

For years, and I say this having both written and reviewed proposals, we treated a well-written document as useful evidence of someone’s capacity. Sure, it might mean that you were a strong proposal writer or had a strong writer on your team, but if you could describe operational detail, it was assumed you probably have some operational experience. If you could name risks, you had probably encountered them. In 2026, we are seeing that heuristic quickly breaking down! An AI system names risks someone has never encountered and describes mitigations they have never tried. In short, text is no longer evidence of having done the work or knowing how to execute what is being proposed.

Some people might wave this off as unethical use of AI or poor use of these platforms. But I think that in a moment when non-profits are struggling to secure funding to continue and trying to make the most with what they have access to, this is a growing phenomenon we need to pay attention to. The gap between what organizations are now (over)proposing and what they can actually deliver is widening as I’ve never seen before and I think it will have substantial ramifications for the nonprofit space. Folks will lose trust in working with new collaborative partners, funders will revert back to their old boys’ clubs and referrals from known friends. I see this is already beginning to happen.

What does it mean for how we select partners when the application can no longer be trusted as a capacity signal?

On the funder evaluation side, it suggests funders begin to discount the proposal and overweight other things: past delivery with receipts not reports, more synchronous meetings and video calls to confirm that a group is who they say they are and build trust in their ability to deliver, a smaller first grant with escalating commitment if the work materializes. But what will happen to open calls in this new world? I noted that the OSV Fund earlier this year received 11,000+ applications and offered 13 scholarships. That kind of application-to-award ratio is unprecedented before this year and will only get more insane.

So many of the dominant critiques about AI are about whether it will take our jobs or how it will make us more productive. But this observation around the growing gap between what is promised and able to be delivered is a bit more nuanced than that…

AI can indeed raise the capacity of organizations and how they present (especially on text-based things like email management; slides and proposals; etc). But AI doesn’t raise the floor of what the humans can do at 6am when something breaks. That second floor is built by practice, by having done it before, by having a team that trusts each other enough to say this isn’t working.

I’ve been working on what exit-to-community and founder extraction at CSIDNet looks like, trying to sort what’s ready to hand off and what is unsafe to hand off. That assessment gets harder when the artifacts of readiness, proposals and decks and plans, can be easily generated (in an instant!) without the underlying readiness actually being there.

I don’t know how to fix this capacity dilemma. We all need more of it but is AI actually offering it or is it shifting the need for capacity to other types/forms. If you’re running a network, funding small organizations, or reviewing applications, I’d love to hear how you’re navigating it.