Serve513 Team

On a service day, Salt Tampa Outreach and Lifespring Church turn a building into a place where neighbors experiencing homelessness come in as guests and move through check-in, clothing, haircuts, and showers. Since September 12, the whole morning has run on Serve513, the app that tracks who is waiting, what they asked for, and which station can take them next.

We at Say Hello Neighbor build that app, and we built it with help from AI. People ask about that, and some are uneasy about it, which is fair. This post explains what the AI did, what it doesn’t do, and where we drew the lines.

AI helped write the software. People run the service day. 3 in 4 changes to the code had AI help. None of the app volunteers use on Saturday runs AI. 57 neighbors served on our first full Saturday.
The short version: AI helped write the software, and people run the service day.

The short version

AI helped us write the software. It is not in the room on Saturday.

Nothing a neighbor or a volunteer uses during a service day talks to an AI. The check-in desk, the queues, and the countdown on the shower stalls are ordinary software that does the same thing every time. No AI decides who gets a haircut, how long a shower lasts, or what goes in a clothing bag. Volunteers make those calls, the same way they always have.

What the AI actually did

The AI is a coding assistant called Claude Code. It works alongside our lead engineer: it writes code when asked, explains its choices, writes tests that prove the code works, and points out problems it finds along the way.

Work on Serve513 started in November 2025. Through September 18, 2026, the AI helped with about three in every four changes to the code: 740 of 991.

Changes to the code each month, with and without Claude. A change counts as with Claude when Claude is its author or co-author. September counts through the 18th.

Its work shows up at every station a neighbor walks through:

  • The Registration Desk, where volunteers sign up new neighbors.
  • The Welcome Desk, where a guest’s needs are written down once and travel with them, so nobody has to explain themselves again at the next station.
  • The barber and shower queues, which update on every volunteer’s screen at the same time.
  • The clothing station, where a guest’s order starts being pulled while they are still at another station.
  • The pickup station, where a finished bag is handed off.
The Welcome Desk, step 2 of 4, asking what Jordan needs today. Showers and Hygiene are selected; Haircuts and Clothing are the other choices.
The Welcome Desk asks once what a neighbor needs today, and every station sees the answer. Shown with the sample profile from our Welcome Desk guide.
The Fill Station. Three clothing orders are in progress with fill bars, and three incoming orders wait: two list their items, one with a Veteran tag, and the third has a note asking for size 11 work boots.
The Fill Station starts pulling a clothing order as soon as a neighbor checks in, while they are still at other stations. Sample data, not real neighbors.
The Pickup Station with one finished bag: 3 items bagged, with buttons for Picked up and Back to fulfillment.
The Pickup Station hands off the finished bag. Sample data, not real neighbors.

Our lead engineer wrote plenty of it by hand too, including some of the biggest pieces. The AI also took on the parts nobody sees: the tests, the written plan behind each feature, and some of the picture guides that walk new volunteers through the desks.

What stayed with people

The best ideas came from watching the floor. On our first Saturday, the showers ran three stalls and took 24 check-ins. The volunteer at the stalls was working out who had run over by checking the wall clock, three stalls at a time, while also calling the next name. You only see a problem like that by standing there.

Once we saw it, the AI helped build the fix within days. Every stall now counts down its minutes, turns yellow when time is nearly up, and changes color again when time runs out.

The Shower Station. Patricia's stall is yellow with 3 minutes left, Diana's stall shows 9 minutes left, two stalls are empty, and four neighbors are waiting, tagged Veteran, Hearing, and Mobility.
Each stall counts down its minutes and turns yellow as time runs short. The tags a neighbor shared at check-in ride along to the station. Sample data, not real neighbors.

Every change also goes through a person before it reaches a service day. Our lead engineer reviews it, and our rule is that nothing merges or goes live without their approval. The AI can suggest, draft, and test. It doesn’t ship anything on its own.

How a change ships, in five steps: the change is drafted, by AI or by hand; automatic lint and formatting checks on every commit; an AI security review on sensitive changes; the lead engineer approves; the change goes live. The approval step is highlighted as the one where a person decides.
The path every change takes. The highlighted step is the one a person owns.

Protecting neighbors’ information

The people we serve trust us with personal details: names, birthdays, clothing sizes, and sometimes things like veteran status or a hearing need. In the weeks before launch, much of the AI-assisted work went into tightening what each role in the app, from station volunteer to team admin, is allowed to see and change.

We also use a second AI reviewer, set up only to check changes to the sensitive parts of the app against our security rules. It flags problems, and our lead engineer decides what to do about them.

The Registration Desk form for a new neighbor named Jordan Avery: first and last name, preferred name, gender, date of birth, phone, email, a Veteran checkbox, and accessibility needs with Mobility checked.
What a neighbor may share when they register. Birthday, phone, and email are optional. Shown with the sample profile from our registration guide.

Why this matters for a small team

Say Hello Neighbor builds Serve513. Salt Tampa Outreach and Lifespring Church host the service days, and volunteers run every station. Before Serve513, the lines at each station ran on paper. For a small team building software for community partners, AI made it realistic to build tools that fit their floor, their stations, and their Saturdays, instead of bending the day around a spreadsheet.

On September 12, the first Saturday run fully on Serve513, the app recorded 57 neighbors registered and served, 59 orders, and 108 station check-ins.

Serve513 soft launch, Saturday, September 12, 2026. 57 neighbors registered and served on our first Saturday. 59 orders, 108 station check-ins, 116 clothing items packed, 3 desk volunteers.
Our first full Saturday on Serve513, from the app's own records.

The software is only there to give volunteers their time back. The work that matters most, a haircut, a hot shower, a coat that fits, and someone who remembers your name, is still done by people.

Come see it

The best way to see a service day is to help run one. Our volunteer guide walks you through signing up. You can read how a service day works on our About page, and see the full numbers from our first Saturday in Soft Launch by the Numbers.