Food Was Never The Problem: What Two Dozen People And One Machine Found In Ninety Minutes

I believe AI is going to amplify who you are or dilute it, and there is no neutral. Which means the tool is never the first question. The human is.

That belief got tested in a restaurant in Phoenix on September 15.

St. Vincent de Paul had invited about two dozen people who fund and support food assistance in that community to dinner, along with the organization's Chief Operating Officer and Chief Philanthropy Officer. The question printed on the invitation was whether AI can help solve food insecurity.

Their Chief Philanthropy Officer, Ryan Corry, opened the evening by telling everybody to keep eating. Plate empty, get up, go get more, because there is plenty. Hold onto that line. He said it again at the end of the night, and it turned out to be the whole point.

Then he handed it to their Chief Operating Officer, who runs almost everything the organization does with food and has been there nearly fifteen years, since she started as a student intern in one of their dining rooms. She spent ten minutes on what she is actually seeing. Arizona has gone from about 900,000 people on food assistance to roughly half that, and her line is the one I carried the rest of the night: they do not necessarily have more hungry people, they have hungrier people.

Then it came to me, and here is the reality: I told them in the first two minutes that I was not going to answer the question on their invitation, and neither was anybody else in the room. I had spent two afternoons preparing verified research and I said so out loud, because presentation can feel fabricated and people can smell it. Then I gave them the only line that mattered. The research is prepped. What is not prepped is you.

What the machine actually is

So we started with the phones. I had everybody pull one out, open whatever AI they already use, and enter the same prompt: give me a random number between one and twenty-five. Almost every person in the room got seventeen. I have run that with a thousand people at once and get the same result.

Then I told it I was a lifelong Michael Jordan fan and asked again. Twenty-three. A NASCAR fan with a favorite driver. Three.

One detail about who is asking changed the answer completely. That is the whole lesson in ninety seconds. It is not intelligent. It is pattern recognition built on historical information, and what you put in front of it decides what you get back. Does that make sense?

What it could not know

Then I opened a brand new chat, attached nothing, loaded no context, and asked it the biggest version of the question on purpose. I did not even type it. I said it out loud:

How do we end hunger for families in Arizona?

It gave us a competent briefing. Give money. Volunteer. Call your legislator. Dial 2-1-1. Good advice for everybody, and that room was not everybody. Ryan pointed out that about seventy percent of those 2-1-1 calls come to St. Vincent de Paul anyway.

It could read their website and their public filings, and the most recent filing predates the entire collapse. It did not know what their COO had just described, and it did not know that one of the people at those tables runs food for one of the largest grocers in the state.

The machine brought the speed. It had no idea where to go until the people in that room told it.

Then the room started talking

So we gave it the room, one person at a time, for about thirty-five minutes.

Near the end I asked the model to tell us what it had that it did not have when I started. Its own answer was that the ideas in that room were not a list. They formed one system: a way to see a family heading for trouble before the crisis lands, a trusted front door where somebody is already standing, and food that is currently being thrown away.

Nobody in that room designed that. A volunteer brought the warning signals, a grocery executive brought the economics and a story about milk, somebody in manufacturing brought stranded inventory, an investor asked for the business case, and a technology founder reframed what the machine is for. Each person had one piece. None of them had the assembly.

Somewhere in the middle of it, I said back to the machine what the room had just worked out, and it is the sentence the whole evening turns on:

There's not necessarily a lack of food. There's enough food to feed all of us. The issue is distribution and access in many capacities.

What I fed the model, from what the room had just said

That is not a small reframe. It moves the problem off scarcity and onto the systems people have to pass through to reach food that already exists. A grocery executive had just described milk poured down a drain because it was packaged in school cartons with no lawful way into gallons. Not a shortage. A system that could not get a thing to the people who needed it.

Complex problems do not get solved by the smartest person in the room. They get solved when enough different kinds of expertise are in the same room and somebody makes sure every piece of it actually gets heard.

What happens next is theirs to decide, not mine to announce. But they left with things to chase that did not exist at six o'clock, and they left holding the one question nobody in the country can currently answer, which is how many of the families who lost food assistance were still eligible the whole time.

What that turned into

Over ninety minutes the model moved from a generic briefing to a specific reading of one organization, one county and one funding decision. Along the way it corrected the COO on one figure, corrected the room on another, refused a question that could not be honestly answered as asked, and when I made it show its arithmetic step by step it caught its own error and cut its projected return in half.

Here is where the room landed. Sixty percent of Arizona's food assistance denials in that period were for a missed interview rather than ineligibility. Only thirteen percent of calls to the state eligibility line reached an interview at all. Which means those families are largely still eligible, and helping one household finish the paperwork returns roughly $3,900 a year in federal food money that is already appropriated and going unclaimed.

Nobody in that room had that at six o'clock. It was not sitting in the machine either. It came from putting them together.

The part I could not have explained until he asked

One thing I left out of all of that. My live transcription tool died about ten minutes in, and the conversation got so good that I cut three blocks I had spent hours building, including one I had promised the room we would reach. So for those thirty-five minutes, nothing reached that machine except through me. Everything the room contributed went in because I listened to it and said it back.

The next morning we debriefed the night. Ryan, who I have worked alongside for twelve years, said the listening and synthesis were unusual, and that the recaps were spot on. Then he made the point that stuck with me. Nobody corrected me all night, and that group was comfortable enough that they would have. He put those two things side by side, my synthesis and their comfort, as what he noticed.

I have been doing it for years without naming it, so his comment is what made me try to explain it. There is no checklist.

The honest answer is presence. Radically transparent about what I am doing and why, and listening in a way I have spent years developing. No notes, no buffer. I am not scanning for items on a list, I am listening for the high point of the arc, and that might be the data, the way the person felt, or the strategy hiding inside a story they think they are telling for color. And it does not work if I am performing. If I am worried about how I am coming across, the whole thing collapses. It requires surrender, letting go of the outcome long enough to actually hear what is in front of me.

The greatest gift we can give someone is our presence. In that room it was also the most useful thing I brought, and it is the part no technology supplies.

Then I realized, none of this was new. It is the same sequence I have used for a decade. Show up human first. Go first myself. Listen until I can say it back accurately. Ask permission before I push. Then tell the truth and hand back something actionable. The machine took a seat at step three, where listening turns into synthesis. I checked my recaps against the recording afterward, and about a third of what was said never made it through me. That is what synthesis is. The skill is not catching everything, it is catching the part the room can build on.

Why this organization, and not just any organization

None of that gets tested without a room to test it in. St. Vincent de Paul is the reason the night existed at all, and that is worth naming.

Most nonprofits are judged on the share of every dollar that goes straight to the cause, which sounds right and quietly caps what any of them can ever do. This organization invests in growth instead, in people and structure and the things that return more than they cost, and they have the receipts. Ryan's own job was that bet once. Twelve years ago they hired a fundraiser to see what he could return, and they have gone from raising about eight million dollars a year to closing in on fifty. That is not inflation. That is structure, the right people, and a willingness to spend money in order to make more of it available to the people they serve.

Which is also why this was their idea and not mine. It started with a phone call a couple of weeks before the dinner, after Ryan and their CEO read an article about someone pushing an AI model past everything it insisted it could not do. They called and asked whether we could do that live, in front of their donors, on the hardest problem they have.

Most organizations do not make that call. Sitting down with an unproven technology in front of two dozen funders, with no guarantee it produces anything, takes the same nerve, and it is why there was anything worth writing about.

There was plenty of food in the room the whole night

Then, at the end of the evening, after ninety minutes on the worst food assistance collapse in the country, Ryan closed the same way he opened. He thanked everyone, and then told them nobody had to leave because there was plenty of food.

I cracked a joke. What a way to end a dinner about food insecurity.

We laughed, and then it sat with me the whole drive home, because it is the entire thing in one sentence. There is plenty of food. There has been plenty of food this whole time. Food was never the problem.

Humans built the system that is failing these families. Which is exactly why humans are the only ones who can fix it, and why the technology's only real job is to compress the time it takes us to figure out how.

That is what came out of a dinner. Not from the technology, and not from me. From two dozen people who were willing to say what they actually know, in a room where it felt safe to do it.

So the question was never what AI can do. It is what you are putting in front of it, whether you were present enough to hear it correctly, and whether you have the courage to find out in public.

Make sure what is already there is worth amplifying. Then, and only then, amplify it.