It Was Already There: Three Stories About the Gap Between Available and Reached
By the GenHax Team · November 20, 2026 · Categories: AI in Education, Knowledge Graphs, Social Impact, Community, Research
This is Hunger & Homelessness Awareness Week, a tradition students started at Villanova in 1975. We spent it on one question that kept coming back in different forms: what happens when the thing people need already exists, and they still can't reach it?
Here's a small experiment. You're in Philadelphia or Los Angeles. It's Sunday. You're hungry, your phone is at 4%, and you have one bus fare. Where do you go?
On Monday we put that question to a knowledge graph. By Friday we'd found the same pattern in an AI tutoring study from Tennessee and a student-built app in San Diego. Each story had a resource that was there the whole time: food, a patient tutor, a tray of leftovers. Each time, the problem was the last few feet between "available" and "reached."
We think that gap is one of the most important and least discussed problems in education right now. Especially in AI education, where the usual answer to every problem is to give people access to more stuff.
Monday: the pantry that might be open
DREAM-KG is a knowledge graph of homelessness and social services built by Huanmei Wu's team at Temple University. It's funded by the National Science Foundation (award #2333703) as part of the Proto-OKN program, and it covers services in Los Angeles and Philadelphia. We queried it directly through the OKN federation.
The graph has 118 distinct services tagged as food pantries. 98 of them list at least one set of opening hours. Only 8 list any hours on a Sunday. 20 list no hours at all.
We want to be careful here. Those numbers don't prove pantries are closed on Sunday. Plenty of them probably open their doors. The data just doesn't say so. But for the person with a dead phone and one bus fare, "the data doesn't say" and "closed" work out the same. Nobody spends their last fare on a maybe.
The scale makes it matter. HUD's 2026 point-in-time count, released November 6, found 698,425 people experiencing homelessness on a single night in January. Each one of them is making decisions with whatever information is in front of them.
The food was there. The information wasn't.
Tuesday: the tutor nobody asked
On Tuesday we read a new working paper by Philip Oreopoulos and Nina Low, "One Click Away: AI Tutoring with Khanmigo in a Two-Year School Experiment" (EdWorkingPaper 26-1551, Annenberg Institute at Brown). It's a cluster randomized trial in 18 Tennessee middle schools in Hamilton County. Students in daily remedial math blocks got Khanmigo, an AI tutor set up to coach students instead of handing them answers.
That's about as good as access gets. It's free, it's patient, it's built into the work, and it's there every day for two years.
Here's what happened:
96% of students tried it at least once.
The median student messaged it on only a third of the days they practiced, and in only 17% of the sessions where they made a mistake.
Only about 1 in 7 messages had a real mathematical question or a step of reasoning. The rest were mostly bare answers (39.4%), clicks on suggested prompts (24.2%), and low-effort messages like "idk" (12.6%).
Achievement went up by about 1.3 national percentile ranks per term. The authors note that's roughly what Khan Academy practice delivers without AI.
That's not a failure. The effect was positive. But it's a long way from the "every child gets a personal tutor" promise, and the authors are clear about why:
Realizing the promise of AI tutoring will require "getting students to use it, not just giving them access."
On the same morning, CodeAI (formerly Code.org) launched its Hour of AI 2026 activities under the theme "Start building." We think that's the right instinct, and it ties straight back to the Tennessee data. A tutor that's one click away doesn't give a student a reason to click. A problem they actually care about might.
The tutor was there. The reason to use it wasn't.
Wednesday: ten minutes
Then we met Daniel Self and Gavin Bartow, San Diego State University Class of '26.
Self, a first-generation computer engineering student, went through community college facing food and housing insecurity. According to SDSU News, he lived out of his car for a time. At SDSU, he and Bartow, an MBA student at the Fowler College of Business, saw something that looks obvious once someone says it out loud. Campus events kept ending with trays of good food headed for the compost. Meanwhile, students down the hall were skipping meals.
They built Second Course. Event organizers post leftover food with a photo, a portion count, a location and a pickup window. Students claim a portion and go get it. The pilot ran on QR-code flyers, and in the beginning the founders posted the food themselves. A $5,000 third-place prize from the 2025 SDSU Social Venture Challenge paid to build the app.
On day one, 150 students signed up. "Nearly every posted meal was claimed within 10 minutes," Self told SDSU News. By spring, the app had 680 users, 15 student organizations, and a partnership with SDSU Catering.
Ten minutes. The food had been sitting there the whole time. Once the information moved, the food moved too.
Bartow said something else that stayed with us: "We want this to be for every student, not just students who are struggling." A good design for reaching people doesn't make anyone raise their hand as needy first. It simply makes the information easy to find.
There's one more detail worth noticing. The person who saw this gap had lived it. That's no accident. The people closest to a problem are usually the first to notice which last step is missing.
What closes the gap: four design moves
None of this week's stories needed more resources. Nobody had to build a new pantry, train a new tutor or cook more food. What was missing every time was the last step. So if you design learning experiences, here's what we're taking from the week.
1. Treat information as infrastructure
The eight Sunday listings aren't a technology problem. They're a data problem, and data problems are exactly what students can work on. Cleaning, linking and checking service data is a real job with real stakes. It teaches knowledge modeling, data quality and empathy all at once. It's also the kind of work that rarely gets funded, because it isn't glamorous.
2. Give learners a problem that's theirs
The Khanmigo study is the clearest evidence we've seen this year that access isn't the same as use. A worksheet doesn't give a student much to ask about. Building something they care about does. When the problem belongs to the learner, "idk" turns into "why does my query return nothing for Sunday?"
3. Put the people closest to the problem in the builder's seat
Daniel Self didn't need a needs assessment to see the gap at SDSU. Last week we wrote about Rachel Ceria, the VA nurse who was invited to mentor at a hackathon and chose to build instead. It's the same pattern. If your program treats affected people only as "users" or "beneficiaries," you're leaving your best problem-finders out of the room.
4. Measure reach, not availability
"96% tried it" sounds like success. "1 in 7 messages was a real question" tells you what actually happened. "118 pantries" sounds like coverage. "8 list Sunday hours" tells you what someone on Sunday can actually use. Whatever you're running, whether it's an AI pilot, a mentoring program or a resource page, find the number that measures the last step and track that one.
Where GenHax fits
We build challenges around exactly this kind of gap.
Our DREAM KG challenge brings the DREAM-KG approach to New York City. Students work on connecting people experiencing homelessness to shelter beds, services and first responders. We've piloted it with 20+ students at Medgar Evers College and Bronx Community College through the NY STEP program. Randell Keiler, NY STEP Program Manager at MEC, put it simply: "GenHax is an easy way for our high school students to experience real-world problems."
That's the point. Students don't learn about information gaps from a slide. They find one in a real graph, argue about how to fix it, and build something that closes a little of it. Marie Francesca Berrouet, Program Manager at C-STEP, Queens Community College, described this kind of work in a way we haven't forgotten: "You just connect the dots for what already existed and made it easier."
Connecting what already existed. That's this week in six words.
For anyone whose students are doing an Hour of AI this December, our Code Your First AI App or Agent challenge is built to be hour two. It takes the "start building" energy and points it at a problem the learner picks.
Your turn
Hunger & Homelessness Awareness Week ends tomorrow. The gap between "available" and "reached" doesn't. Somewhere on your campus or in your company, something useful is one click away and nobody is clicking. Somebody already knows why.
Find that person. Hand them the problem.
Browse live challenges, including DREAM KG and Code Your First AI App or Agent, at app.genhax.com/explore. Educators and program managers can see how we run challenges with colleges at app.genhax.com/academia.
The next issue of our newsletter, GenHax Chronicles, comes out Monday. Subscribe on LinkedIn to get the fortnight's best research and challenges in one read.
Sources
DREAM-KG (Temple University; PI Huanmei Wu; NSF award #2333703, Proto-OKN). Live SPARQL query via the OKN federation, Nov 16, 2026. These are listed opening hours, not verified ones. registry.okn.us
U.S. Department of Housing and Urban Development, press release No. 26-090, Nov 6, 2026. hud.gov
Hunger & Homelessness Awareness Week, Nov 15 to 21, 2026 (National Coalition for the Homeless; National Student Campaign Against Hunger and Homelessness). hhweek.org
Philip Oreopoulos & Nina Low, "One Click Away: AI Tutoring with Khanmigo in a Two-Year School Experiment," EdWorkingPaper 26-1551, Annenberg Institute at Brown University (working paper). doi.org/10.26300/kner-hv33
CodeAI, "Start Building: Hour of AI Gives Students a Hands-On First Step Into Understanding Technology," PR Newswire, Nov 17, 2026. prnewswire.com · hourofai.org
Peggy Pico, "Class of '26: Graduating seniors develop app targeting food waste; students eat it up," SDSU News, May 8, 2026. sdsu.edu
VA Pacific Islands Health Care, "Rachel Ceria's 8,500-mile journey to 2025 Veterans Health Hackathon," Nov 25, 2025. va.gov
GenHax: DREAM KG pilot details, app.genhax.com/academia; community testimonials from genhax.com.