TL;DR: If I had $50M for AI resilience, I'd spend it on the people who work directly with young people: youth housing providers, schools, and community organizations. They lose a huge amount of time to paperwork, and a lot of that paperwork comes from funders and public agencies. I'd build a three-year fund that gives these organizations AI tools that fit how they actually work, goes after the paperwork where it starts, and tests whether any of this beats simply giving them cash.
One afternoon
Early in my career, before I moved into grants and systems work, I wrote a lot of case notes. So let me start there.
A case manager wraps up a meeting with a 17-year-old who's about to lose their housing. That hour was the part that matters. She listened, earned a little more trust, and figured out what this kid needs this week. Then she sits down to document it. By the time she's done, the afternoon's gone, and so is the call to the landlord that might have stopped the eviction notice.
The case note is only one piece. There's intake paperwork, eligibility checks, referrals, follow-ups, and billing. And there's the part anyone who's done this work will recognize: typing the same young person's story into five different systems for five different funders. The agency's own records. The federal homeless data system. A state contract portal. Two foundation reports, each with its own format and its own deadline.
Now imagine she talks through the meeting, and AI drafts the note, fills in the intake fields, and pre-fills the reports. She reads it over, fixes the one thing it got wrong, and signs. The time she gets back goes to the landlord, or the school counselor, or the next kid on her list.
That's the part I care about. The relationship still belongs to the case manager. AI just clears space for it. For a young person on the edge of homelessness, that can mean the call gets made today instead of next week. For the case manager, it can be the difference between staying in this job and burning out.
Where the paperwork comes from
I've seen this paperwork from three sides.
Early on, working directly with people, I learned that every hour of documentation is an hour not spent with someone.
At the state level, I managed a federal grant that ran through more than ten provider agencies. I trained their staff and audited their client files. A lot of what I was checking was there because someone above them asked for it: a federal rule, a state contract, a funder's report template.
Now I'm on the funder side. I run grants operations at a foundation and lead our AI adoption. I've built AI tools that took hour-long grant tasks down to 10 or 15 minutes. I've also seen the requirements we send out to grantees, and I know what they turn into on the other end.
Put those three views together and the pattern is hard to miss. Paperwork rolls downhill. Funders and agencies create it, and the people closest to young people absorb it. Making paperwork faster without asking for less is like buying a bigger bucket for a leaky roof. It helps for a while, and then the requirements grow to fill the time you saved.
There's an upside hiding in here too. Frontline records are where a community's data starts. If those records are quicker, more complete, and entered once instead of five times, communities get better information to match young people with housing and services and to catch problems earlier.
Why this counts as AI resilience
The Foundation's definition of AI resilience includes AI's impact on young people and protecting institutions that don't have many resources. Youth-serving frontline organizations fit both. And they're going to use AI whether anyone helps them or not. Without help, the most sensitive information about the most vulnerable kids ends up in tools nobody checked, and an AI-written record with a mistake in it can follow a young person into court or a child welfare file. Helping these organizations adopt AI safely lowers that risk and makes sure the benefits actually reach them.
How the fund would work
I wouldn't build a custom system for each organization. That gets expensive, small organizations can't maintain it, and you can't compare results. I also wouldn't build a toolkit ahead of time and hand it out, because it would be designed for an organization that doesn't exist. I'd build the tools alongside the first group of grantees.
For one organization, it goes like this. They join the pilot, and a small team sits down with their staff to figure out where the hours really go. What we learn across all the organizations shapes a shared set of tools for the things almost everyone deals with: documentation, intake, entering the same data over and over, and reporting. Then the team fits those tools to that organization's systems and way of working. When something works for one organization, it goes back into the shared toolkit so everyone gets it. Everyone follows the same rules for protecting young people's information.
The money
About $27M for AI access and the shared tools. This covers model access and AI capacity for a first group of about 20 youth housing providers, schools, and community organizations in a few communities, plus building and maintaining the tools. Resources go out in stages, based on actual use.
About $10M for the people doing the hands-on work. That means the time studies, fitting the tools, and training staff. The training is about using AI to back up professional judgment, not replace it. The ground rules include privacy and consent practices, secure setups, and a person reviewing every record before it's final.
About $6M to go after the paperwork where it starts. This part works with funders and public agencies in each pilot community:
- It would publish a map of how many staff hours each requirement actually costs.
- It would test a "report once" approach: participating funders agree on one core report, and AI converts it into each funder's format while everyone adjusts.
- It would test ways for information entered once to reach the systems that need it, with consent.
- The Foundation would go first by simplifying what it asks of its own grantees.
About $7M to find out if it works. We'd compare three groups: organizations that get AI resources and support, organizations that get the same value in cash, and organizations that get both. We'd track how many hours go back to direct service, staff turnover, how accurate the records are, and outcomes for young people, like staying housed and staying in school. Results would be published every year.
Over three years:
- Year one is the time studies, the first shared tools, and early results.
- Year two grows what worked and drops what didn't.
- Year three helps grantees plan for life after the fund and publishes a full report and a playbook other funders and agencies can use.
Why not just give cash?
Cash is flexible, but in this field it mostly buys more staff hours, hiring is slow, and burnout is high. AI can give time back to the people already there. Tools alone don't do it either, because without someone helping, they sit unused. And tools without fewer requirements just get swallowed by new paperwork. That's why the fund does all three.
What could go wrong
The risk that worries me most is mistakes in records. A wrong detail in an AI-drafted note can hurt a young person. So a person reviews every record, and the fund spot-checks AI-drafted work the same way I used to audit provider files.
Privacy is a close second. These are records about minors and people in crisis, so the fund needs strict data agreements, secure setups, and tighter access for the most sensitive work.
Some organizations won't take to the tools. That's why resources go out in stages, support is built in, and unused resources move to organizations that will use them.
Some funders won't change. The hours map puts a real cost on their requirements, and the ask is easier when the Foundation has already gone first.
Tools can also end up built for communities instead of with them. Frontline staff and young people should help design and oversee the tools in every pilot community.
What success looks like
After three years, I'd want to be able to say:
- Frontline staff in the pilot communities spend noticeably more of their week with young people.
- There's a shared, open set of tools any frontline organization can pick up, with safeguards built in.
- The hours map is public, and funders in the pilot communities accept one common report.
- We have a real answer to whether AI resources beat cash for these organizations, and a playbook for doing it safely.
Why I think this way
I started volunteering with nonprofits when I was young, and that's what led me to get a master's in human services. My career has moved from the ground up: early frontline work with individuals and families, where I wrote case notes of my own, then running a statewide federal grant, and now philanthropy at the systems level. I've been the one writing the case note, the one auditing it, and the one funding the work behind it. So when I think about AI, I think about both levels at once. The big picture only matters if it changes what happens in the room between a case manager and a kid who needs help.
Morgan Keys runs grants operations and leads AI adoption at a private family foundation in Seattle. These views are personal and don't represent any employer.