We let agents
run the back office.
Humans run the conversation.
A non-profit community has the same admin load as a small company, with none of the budget. So we automate the parts that can be automated — and put every minute we save into the parts that can't.
· Agents handle what the org does.
· Members decide why, and for whom.
· Every agent output gets a human checkpoint before it leaves the building.
· If automation makes the relationship thinner, we don't ship it.
The admin never stops.
Five problems that used to eat volunteers' evenings.
→ Hover or tap a card — meet the agent.
Things we'll never
hand to an agent.
The agents free up our time so we can spend it here. If a process shows up on this list, it doesn't matter how repetitive it looks — a real person reads, replies, decides.
Member feedback & project critique
Every piece of feedback from a member is read by a human. No summarization, no auto-tagging, no AI in the loop.
1-on-1s and conflict resolution
Hard conversations are the work. Outsourcing them to an agent would be outsourcing the relationship itself.
Decisions about specific people
Tier promotions, role assignments, anything that touches a member's standing in the community — humans only.
Judging pledge violations
The pledge is the social contract. Calling someone in on it is something only a member can do, to another member.
Creative direction reviews
When a project decides what it is and isn't, that's a conversation between humans who care. AI doesn't get a vote there.
Why we built it this way.
↳ The argument for letting agents in — and the line we won't cross.
Non-profits live in a bind. The mission demands a small-company workload — accounting, comms, partnerships, member ops — but the budget is zero, the staff is volunteer, and the cost of asking someone to spend their evening formatting expense reports is real human burnout.
The traditional answer is to do less. Run fewer events. Reply slower. Skip the partner outreach. Let the newsletter slip. We have watched good communities die this way — not from a lack of passion, but from the quiet attrition of people who joined to build things and ended up doing inbox triage for a year.
We think there is a third option now. Most of the admin load — the part that is repetitive, structured, and rule-following — can be drafted by an agent. The judgment, the relationships, the actual decisions stay with members. The same volunteer who used to spend three hours on reimbursements now spends fifteen minutes reviewing an agent's draft, and gets the other two hours back — to talk to a new member, ship a project, or just rest.
The agents are interns who never sleep and never get bored. We're the ones who decide what they're allowed to ship.
This isn't about pretending humans aren't there. Every agent output crosses a human checkpoint before it leaves the building. The drafts are cheap; the review is the work. If an agent draft would embarrass us in front of a member, it doesn't go out — and the next prompt gets tightened.
What we are trying to find out is exactly where the line goes. There is a healthy version of this where the org runs leaner, members spend more time on what they came for, and nobody loses the connection to how decisions get made. There is also a bad version where automation makes the relationship thinner and the community becomes a brand instead of a place. We are betting we can find the first one. This page is the public record of how we are trying.
Help us figure out where the line should go.
Every new project teaches us something about what to automate and what to keep slow and human. Join the community — bring an opinion.