Hypothesis snapshot, 2026-04-25.

What GiveReady proposed to the Gates Foundation, captured the day the application was submitted. Nothing below has been edited since.

Frozen snapshot, 2026-04-25

This page is a static record of what GiveReady claimed at the moment the Gates Foundation application went in. It is not the current state of the project. The point of freezing it is so any reviewer can compare what was proposed against what the data has since shown, in our own words on both sides. The current state lives at /progress. The first iteration since this submission lives at /progress/2026-05-24-pivot.

Source: GiveReady-Gates-Proposal-v6.pdf, submitted 2026-04-25 to the AI to Accelerate Charitable Giving Grand Challenge.

The problem we were solving

Charitable giving is heavily concentrated. The largest US public charities receive the overwhelming majority of private contributions: roughly 85% of donations flow to the top 10% of organisations. The remaining 1.5 million small and mid-sized charities, the long tail that does most of the direct community work, compete for what is left.

AI assistants are becoming how people discover where to give, and without intervention they will amplify that power law: when an agent answers "who should I support?", it answers from the thin subset of charities whose websites, registrations, and mission statements happen to be legible to large language models. Small charities are unlisted, unscraped, and unreachable. Giving intent dies in the gap between asking and acting.

The application addressed Challenge Area 3 (foundational data systems interoperable with AI agents) directly, while enabling Area 1 (connecting donors with causes through AI-mediated discovery) and Area 2 (turning interest into completed donations).

What GiveReady proposed to be

GiveReady is open, AI-native infrastructure that lets AI assistants match donor intent to verified nonprofits and complete donations in one step, so donors give more and give sooner, and so the long tail of small charities becomes discoverable and fundable, not just the top of the pyramid.

Two donation paths: x402 (sub-second USDC settlement, no intermediary, no platform fee) for charities with crypto wallets; a fiat gateway path for charities without wallets (an agent pays GiveReady, we remit to the charity's bank account via standard fiat rails, and the charity sees "Donation via GiveReady" arrive on their statement). Two enrichment paths: AI assistants pull programme details, beneficiary demographics, and impact data from public sources and submit as structured context; human reviewers validate every enrichment against regulator registries as ground truth.

The five theories the application rested on

The proposal did not number these explicitly. Numbering added 2026-05-23 for tracking. Wording on this page matches what was implicit in the v6 text.

  1. AI agents will discover and act on small-nonprofit data when it is exposed via MCP and AGENTS.md.
    When agents query on behalf of donors, they also enrich. Multiple independent agents submitting the same value for the same field auto-promotes it; disagreement returns the rejection reason as prior context for the next enrichment. The grant rested on agents both reading the data and writing back to it.
  2. Donors actually want to give via AI assistants.
    The primary outcome metric was AI-mediated donations completed. The proposal assumed that once the infrastructure was in place, donor intent expressed to an AI assistant would convert into completed gifts.
  3. The gateway payment path works end-to-end for charities without wallets.
    An agent pays GiveReady, GiveReady remits to the charity bank account via standard fiat processor rails, the charity sees "Donation via GiveReady" on their statement and comes to claim their profile afterwards. The whole cycle was proposed as deployable in Months 1-3.
  4. The framework holds across UK, US, and South African regulatory environments.
    Three jurisdictions tested in parallel. City Kids Surfing as the named UK pilot. Joe Taylor as UK Charity Sector Lead with ACEVO, Charity Finance Group, Institute of Fundraising, School for Social Entrepreneurs networks. US reached through the IRS 990 registry. A South African youth-charity cohort selected in Months 1-2. The claim was that if the framework holds across these three, it holds for most small charities worldwide.
  5. B2A scales at zero marginal cost.
    Every AI assistant that indexes the MCP registry was proposed to become both a distribution partner and a data contributor at zero marginal cost. Infrastructure runs at approximately $50/year on Cloudflare. Path to sustained impact rests on this: scaling does not require additional funding or nonprofit recruitment.

What we said we would build, by month

Months 1-3
Deploy the gateway path. Validate AI-assisted enrichment on 1,000 nonprofits across the UK, US, and South Africa. Onboard City Kids Surfing as the first fully verified pilot with organisational wallet and governance toolkit (Squads Protocol multi-sig custody, Bitwave audit trails, trustee-liability guidance).
Months 4-8
Run 200 to 300 live AI-initiated donations across both payment paths. Month 3 and Month 6 checkpoint reports go to the Foundation. If AI-initiated donations are near zero at Month 6, the hypothesis needs revision and we will report findings transparently.
Months 9-12
Expand toward 5,000 to 10,000 nonprofits. Publish the open-source governance toolkit and enrichment methodology under MIT licence, as a reusable toolkit other funders, donor-advised funds, and platforms can adopt or challenge without rebuilding it.

Donor-behaviour indicators we said we would track

From the proposal, in the RFP's terms. These were the success metrics at submission.

What was already live at submission

From the proposal: a working prototype was already deployed and in active use. Live API with 41,000 nonprofits on Cloudflare Workers. MCP server on the official registry. x402 donate endpoint tested. Open-source donate widget running on a live nonprofit site. AI discovery routes (llms.txt, agents.md, ai-plugin.json) already receiving named-agent traffic from Claude-SearchBot, Amzn-SearchBot, and MCPRegistry-Crawler. The infrastructure was real on day one; the question the grant was asking was whether the behaviours we proposed would actually happen on top of it.

Consortium named in the application

TestVentures.net led technical development and served as lead applicant. Joe Taylor, a City Kids Surfing trustee with two decades inside the UK charity sector (ACEVO, Charity Finance Group, Institute of Fundraising, School for Social Entrepreneurs), served as UK Charity Sector Lead, responsible for introductions, governance advisory, and trustee-facing credibility. City Kids Surfing (UK) was the named UK pilot charity. US nonprofits were to be reached through the IRS 990 registry. A South African pilot cohort was to be selected in Months 1-2.

What this snapshot is for

A grant application is a hypothesis. Some of the hypothesis will hold up against live data. Some of it will not. The point of leaving this page frozen, exactly as submitted, is that any later page on this site (including the current /progress view) can be compared to this one without anyone editing the past. The next snapshot after this one was 2026-05-24, when the autonomous-agent-write half of theory 1 and theory 5 was retired based on six weeks of live data. Read it at /progress/2026-05-24-pivot.

← Back to /progress Next snapshot: 2026-05-24 pivot →