Case Study · AI-Managed Treasury · Non-Profit
A $10M reserve, managed with an agent stack.
A national non-profit (501(c)(3)) held its reserves in long-term investments and low-yield money market. Over a nine-month engagement we designed a long-horizon 80/10/10 reserve mandate, built a three-layer AI agent stack to run it — agents reviewing agents, humans in the loop every day — and handed the whole system to their internal team. The fund is live, running at a $10M start.
The organization
Reserves working at money-market speed.
A national non-profit with real reserves and a real mission — and a treasury doing what most non-profit treasuries do: sitting in a mix of long-term investments and low-yield money market, safe but stagnant, while the mission it exists to fund moved faster than the money behind it.
The board didn’t want a trading desk. They wanted a durable, transparent, long-horizon reserve strategy — one that a board could govern, an auditor could follow, and a small team could actually operate. That last part is where AI came in.
The mandate · as of July 1, 2026
The reserve, at a glance.
Earnings to date — the yield sleeve
Cumulative yield on the $1M USDC sleeve at its stated 6% APY: ~$5K/month, ~$45K by July 1. This is arithmetic on the mandate’s stated terms — not a reproduction of the client’s statements. The BTC sleeve carries no chart here on purpose: a long-horizon, no-active-trading mandate doesn’t narrate itself month to month.
Allocation
The agent stack
What we did
Nine months, six moves.
Designed the mandate
An 80/10/10 structure built for a board, not a trader: 80% Bitcoin held long-horizon with no active trading — volatility absorbed by time horizon, not by trading — 10% USDC earning a stated 6%, and 10% deployed as active giving. Generosity isn’t what’s left over; it’s an allocation.
Paid the tuition ourselves
Before this engagement, we ran our own automated trading experiments in public — with our own money — and published the autopsy when a strategy failed. Every lesson that survived — position discipline, kill-switches, monitoring that never sleeps, knowing when a market is telling you no — was incorporated into this stack. The client got the lessons without paying for them.
Built the three-layer agent stack
Monitoring, rebalancing & risk, reporting — each its own layer, each with its own agents, an orchestrator coordinating above them. We forked the RAG and personality of our own in-house analyst agent as the foundation, then tuned it to a treasury’s temperament: patient, skeptical, allergic to excitement.
✓ Live in productionMade the agents watch each other
Agents review other agents’ work before it reaches a human. A monitoring alert gets checked by a second agent before escalation; a rebalancing proposal gets an independent risk read. Disagreement escalates to people. No single model’s output moves money.
Kept humans in the loop — daily
The same governance rule we ship everywhere: agents notify and propose; they do not act. The client’s team reviews the stack’s output every day. A long-horizon, no-active-trading mandate makes this easy — the system’s job is vigilance, not velocity.
Handed it off
The stack is run today by the client’s own operation: two fund managers, a controller, board oversight, and two AI engineers. Our engagement is closed. That’s the point — we build systems clients own, not dependencies they rent.
✓ Engagement closedWhere it stands
Live, governed, and theirs.
Why it worked
Horizon beats activity
A long-horizon, no-active-trading mandate is one a board can govern and a small team can run. Vigilance, not velocity.
Agents review agents
No single model’s output moves money. Independent agent review first, human review always, disagreement escalates.
Tuition already paid
The risk playbook came from our own public trading experiments — including the one that failed. Scars transfer; losses don’t have to.
Build owners, not dependents
Nine months, then handoff to the client’s own team. The engagement ends; the system stays. That’s the model.
Reserves working at money-market speed?
We design governed, long-horizon treasury systems that a board can trust and your own team can run — and we hand you the keys.
Start a conversation → More case studies