Last week we introduced the six organizations testing agentic giving with us. One of the questions on their list: how do we make sure our cause is the first answer when someone asks AI where to give?
That question stops one step short; once the agent has the answer, it has to pay.
đ Thought for the week
On August 14, Aashis Luitel of the University of the Cumberlands published a piece in The Conversation about that step. His example is a small one: an agent told just to search for a shirt, buys one instead.
What happens next: The retailer logs a sale, the processor logs a charge, and the AI provider logs a session. Every one of those records is accurate, and as Luitel puts it, ânothing in those records links the charge to the task.â
Thereâs no evidence chain running from what the person asked for to what the agent did, and because the systems belong to different companies, nobody holds enough of the picture to settle the argument.
Washington has noticed and is putting in place things to be aware of. On July 21, Senator Mark Warner introduced the AI AGENT Act (S. 5051). It defines a âcustodial user agentâ as one authorized to act for a user in a transparent, documented, limited and revocable way, and it hands NIST the job of identifying protocols or writing standards for verifying that a user delegated authority in the first place. Google has published its Agent Payments Protocol, AP2, which covers some of the same ground by creating records of the limits a user approved and the information each party saw when a transaction is disputed.
Neither is finished. AP2 produces the evidence but doesnât say who takes the loss, or how long anyone has to keep the records. And NISTâs current work is scoped to agents operating inside a single organization. The consumer agent that crosses company boundaries, which is the one your donor will actually use, has been pushed to later.
A gift is the easiest version of this problem
When a retailer canât prove authorization, it has a dispute with a stranger over a shirt. When you canât, you have a conversation with someone who already chose you.
Thatâs a far better place to be standing. The amounts are smaller, the relationship is longer, and unlike almost every other transaction on the internet, both sides want the record to exist. Your supporter wants their gift recognized properly and you want to know what they actually asked for. Nobody is trying to hide anything from anybody.
So the sector has a reason to reach this before commerce does rather than after. The standards are being written right now, in a bill and a protocol that are both still open, and nobody has asked what a donation looks like inside them. Nonprofit payment processors will move slowly, which means the useful move isnât waiting for them. Itâs keeping your own record of what a supporterâs agent asked for, in your own CRM, starting with the first agentic gift you receive. Something weâll be testing in the cohort.
The part worth being cheerful about
Everything on Luitelâs list, a task identifier that travels between systems, stated limits, tamper-evident logs, adds up to one thing; a record of what a supporter actually asked for.
Thatâs the most useful consent and preference data our sector has ever had access to. Fortune ran a piece on Sunday on the $124 trillion wealth transfer running through 2048, quoting Bloomerangâs Steve Isom on what younger donors want. 94% say theyâre motivated by an organization being specific about where the money goes, and specific impact language beat generic messaging by 88 percentage points. An agent can only pass on specificity youâve already published. Write it down, build to receive a structured record back, and your next conversation starts with a segment of one instead of a guess.
That is more or less what the six organizations in our cohort spend the next five months on. When the first of these records starts arriving, weâll publish what it actually looks like.
Three things you can do this week
Send your payment processor one email. Whatâs your roadmap for agent-initiated payments, and are you looking at AP2? If nothing, join our cohort two waitlist so you have something practical by the end of the year.
Decide how youâd handle a supporter who queries a gift their agent made. For most of you the answer is refund it and ask what happened, which is also the answer that keeps the supporter.
Look at what your site tells a machine. Program names, ask amounts, what a given amount does. Thatâs the material an agent quotes when someone asks it where to give.
đ Join the second cohort
The waitlist for Cohort 2 is here: Join the Cohort 2 waitlist
Same shape as the first. Five months, real deployments, findings published here as they happen, failures reported to learn from as well as the wins. If youâre sitting on a dormant list, a large pledge file, or an audience youâve never asked, thatâs the profile weâd like to hear from.
âď¸ Weâre in Chicago
Weâll be in Chicago from 1st through 5th September, presenting with our partners at Project Equity.
If your organization is in or near Chicago and you want to talk properly about any of this, the cohort, agentic giving, or the state of your own supporter data, reply to this email and weâll find a time. Weâd rather spend the week in conversations than in sessions.
đ¤ Funding radar
Miller Center for Global Impact accelerator (Santa Clara University). A free accelerator for social enterprises working on womenâs economic power or climate resilience. Early deadline August 31, final deadline October 16, for the June 2027 cohort. Details
JD Finish Line Foundation, Louder Than Words Cycle 3: Safe Communities. Up to $10,000 for nonprofits serving underserved communities nationwide. The portal opens September 1 and closes October 15. Details
We Raise Foundation Emerging Leader Grants. $15,000 over two years, split between $10,000 of program funding and $5,000 of professional development, for early-career leaders at Christian 501(c)(3) organizations. Deadline September 30. Details
Worth a look
The $124 trillion handover (Fortune, Aug 16). 75% of millennials say they plan to increase their giving year over year, and about 2% of households hold more than half the wealth in transit. The line worth stealing is Bloomerangâs: donors are ready to trust nonprofits, they just want to see the receipts.
Whoâs building AI for good (EY). Women lead half of impact enterprises but only a quarter of AI-driven ones, and women-led startups take 2% of global venture funding. Read it next to your own vendor list.
An AI agent spent your money (The Conversation, Aug 14). The full piece behind this weekâs lead, including the five components Luitel argues any accountable agent payment needs.
Reply to this email if you want to meet in Chicago, or if you have a view on any of the above. We read every one.