The Coordination Stack

Governments and multilaterals move $200B+ annually through development channels with 40-60% leakage. No open-source treasury management system exists that combines programmable allocation rules, milestone-based disbursement, real-time audit trails, and stablecoin settlement.

Outcomes-based financing (impact bonds, retroactive funding, hypercert buybacks) is proven in concept but has no shared infrastructure. Every implementation is bespoke. UBS Optimus raised $100M for outcomes financing but the rails to deploy it efficiently don’t exist.

$600M in new climate instruments were structured in early 2026 for emerging markets, but deploying capital requires verified outcome data that traditional manual reporting cannot provide at scale. No open protocol connects digital measurement, reporting, and verification (dMRV) data to programmable disbursement rails, and the need extends beyond climate to education, health, livelihoods, and infrastructure outcomes.

Every protocol on this stack settles in USDC or USDT by default, meaning the entire coordination layer has a single point of capture at the money layer. One company can freeze any wallet without judicial process. Sky (formerly MakerDAO, now issuing USDS in place of DAI) proved decentralized stablecoins could work but not scale. Terra proved they could scale but not survive. The GENIUS Act (July 2025) codified centralized USD stablecoins as a regulated category, which sharpens rather than weakens the case for a decentralized alternative: regulatory capture is now the explicit risk vector. No stablecoin exists that is decentralized in issuance, governance, and collateral, with a multi-jurisdictional collateral base that prevents any single government’s bond market from becoming the control vector.

The global financial system offers a false binary: state-controlled rails (SWIFT, CIPS, BRICS Pay) or private-sector controlled rails (USDC, USDT). Both concentrate power. Both can be weaponized. No open-source payment infrastructure exists that settles natively in multiple CBDCs and sovereign currencies, complies with any jurisdiction’s laws through node-level ZK compliance verification, and remains resistant to unilateral capture by any single state or corporation. The neutral interoperability layer between sovereign monetary systems, without the geopolitical strings of SWIFT or BRICS Pay.

Every institution deploying crypto rails needs KYC/AML compliance. Current solutions require centralizing personal data with intermediaries. ZK proofs can verify compliance without exposing underlying data. Production attempts exist (Bandit, Nocturne, Privado ID; Sismo wound down in 2024), but none has achieved institutional adoption at scale. The gap is no longer technical proof-of-concept; it is the integration, audit, and procurement work required to make ZK compliance the default rather than the exception.

The Maduro Paradox demonstrated that stablecoins extend rather than escape dollar surveillance. Privacy coins have liquidity and regulatory problems. No production-ready private payment system exists that satisfies both user privacy and regulatory compliance through selective disclosure.

Moving funds from international donors to local beneficiaries requires 4-7 intermediaries, each taking a cut and adding weeks of delay. No end-to-end system connects institutional senders to last-mile recipients with the compliance, verification, and reporting that institutions require.

The hardest problem in development finance and sovereign coordination is the final mile: moving from crypto rails into local banking rails instantly and compliantly. Institutional buyers will not hold crypto on their balance sheets. No product provides programmatic FX and compliance API layers to auto-liquidate agent-driven stablecoin payments directly into sovereign fiat systems (M-Pesa, Pix, UPI, local bank accounts).

India committed $200B+ to sovereign AI and has provisioned over 58,000 sovereign GPUs through the IndiaAI Mission (34,000 in the initial tranche plus 20,000+ pledged at the February 2026 India AI Impact Summit), with a 100,000 GPU target by end of 2026. Utilization on the initial tranche ran around 22%. Enterprises still choose hyperscalers because VCs bundle cloud credits and the ecosystem is stickier. Building sovereign infrastructure doesn’t mean anyone uses it. The missing piece is the middleware that makes sovereign compute competitive.

Sovereign AI buildouts assume trust in the compute provider. Ministries running inference need guarantees that data isn’t leaking and computations ran correctly. TEEs provide hardware-attested confidential execution; SNARK-based verifiable compute has dropped overhead by orders of magnitude over the past two years and is approaching viable institutional economics. No product packages these as an institutional verification layer for sovereign compute, agent execution, or multilateral data workflows.

100+ countries are provisioning sovereign GPU clusters, creating localized gluts and deficits in compute power (India’s 22% utilization rate is typical). No marketplace exists for mid-sized economies to trade compute cycles, sell forward contracts to global AI labs, or hedge their massive infrastructure costs.

Sovereign states will adopt open-source coordination protocols but will reject public token-gas models controlled by anonymous validators. No shared, permissioned consensus infrastructure exists for allied sovereign blocs that settles in CBDCs or authorized stablecoins while maintaining decentralized coordination.

As AI capabilities become powerful enough to threaten state sovereignty, the entities controlling them are already seeking jurisdictions beyond state reach. Tether operates from the BVI. Crypto clusters in Dubai and Singapore. Starlink controls communications for 100+ countries from satellites no single nation regulates. Coordination infrastructure designed to operate within existing legal frameworks will not survive this trajectory. The coordination layer must function regardless of where the compute, the operator, and the user are located.

As AI agents enter economic and governance systems, the foundational question becomes: is this entity human, and if not, what is it authorized to do? No decentralized, privacy-preserving system exists that handles both human and agent identity.

World (Sam Altman’s project) has verified nearly 18 million humans via centralized biometric hardware linked to OpenAI, with a stated target of 100 million within 12 months. It solves the right problem but through centralized capture. If World ID becomes the default identity layer for the agent economy, it creates a new concentration risk.

AI agents are transacting autonomously with no governance framework. What can an agent do? With whose resources? Under what constraints? Who’s liable when it fails? No protocol exists for defining, enforcing, and auditing agent permissions in economic contexts.

Anthropic’s Constitutional AI approach (training models against a set of principles) has no equivalent for governance systems. No framework exists for encoding community values into agent governance protocols in a way that is auditable, updatable, and resistant to capture.

Agents need to discover, negotiate with, transact with, and resolve disputes with other agents at machine speed. No marketplace exists where trust is programmable, accountability is verifiable, and the governance layer prevents autonomous systems from concentrating power.

Eight competing agent payment protocols launched within a 90-day window in early 2026: Visa TAP, Google AP2, Coinbase x402, Stripe/OpenAI ACP, PayPal Agent Ready, Alipay’s Agentic Commerce Trust Protocol (ACTP), Google and Shopify’s Universal Commerce Protocol (UCP, launched January 11, 2026), and Stripe and Tempo’s Machine Payments Protocol (MPP, launched March 18, 2026). No neutral interoperability and governance layer exists between them. Settlement concentrates through USDC on the crypto-rail side (98.6% of on-chain agent transactions). Middleware bridging crypto-rail agent commerce (x402, AP2, MPP) with fiat-rail agent commerce (ACTP, UCP, UPI Circle, Pix agent flows) is currently unfilled and spans incompatible regulatory regimes.

When multi-agent ensembles (Scout, Analyst, Outreach, Payments, Memory in typical ReAct architectures) collaborate on a task, no protocol exists for attributing value back to each contributing model, dataset, or sub-agent. a16z’s Scott Kominers flagged this in January 2026 (“operating ensembles of wrapped reasoning agents will require better interoperability between models, along with a way to recognize and properly compensate each model’s contribution”). The settlement layer: micropayment routing weighted by verifiable contribution. Without an attribution primitive, the agent economy will default to winner-take-all model-provider economics, the rent-extraction pattern Commons Lab exists to counter.

If an autonomous agent executes a trade, misallocates funds, or causes harm, liability must attach to a legal entity. No automated infrastructure exists for incorporating, tax-routing, and generating liability shields for autonomous agents. Existing incorporation services (Stripe Atlas, Clerky) are designed for human founders. Creating legal wrappers manually does not scale to millions of agents.

As autonomous agents transact at scale, the financial risk of hallucinations, logic errors, and adversarial exploits scales with them. No insurance product exists that algorithmically underwrites autonomous agent risk, escrows liability funds, and settles claims when agents cause harm.

The agent economy assumes pre-funded wallets, but locking stablecoin liquidity across thousands of individual agent wallets is massively capital-inefficient. Corporate treasuries want just-in-time credit, not floating balances in every agent. No under-collateralized lending primitive exists for machine agents.

When an autonomous agent breaches an SLA, hallucinates a catastrophic trade, or misallocates funds, traditional legal recourse is impossible if the deployer is anonymous or in a non-extradition jurisdiction. No product exists that enforces economic accountability for autonomous agent behavior without relying on the legacy legal system.

Chainalysis and Elliptic detect illicit blockchain transactions at human investigation speed. The agent economy operates at machine speed. Flash-loan cascades, multi-agent coordination attacks, and automated capital flight execute in milliseconds. No product exists that detects and responds to illicit autonomous agent activity in real time.

Governments spending billions on sovereign AI need absolute certainty that the models they deploy haven’t been backdoored, poisoned, or tampered with. No product exists that provides cryptographic verification of model weight integrity, training data provenance, and supply chain auditability for frontier models deployed on sovereign infrastructure.

Glasswing proved that frontier AI models can find thousands of zero-day vulnerabilities across all major software. Anthropic released Claude Mythos under Project Glasswing only to a coalition of launch partners (Apple, Google, Microsoft, AWS, NVIDIA, JPMorgan Chase, Linux Foundation, others) plus around 40 additional critical-infrastructure organizations. Disclosure norms have since loosened so partners can share findings with regulators, OSS maintainers, and the press. But concentrated access to working exploits for most of the world’s critical software still sits inside one private company. No mechanism exists to distribute defensive security capability beyond the invited coalition without centralizing offensive knowledge in the process.

Tailwind CSS lost 80% of revenue despite 75M monthly npm downloads and 51% developer adoption. AI coding tools generate code directly, bypassing documentation and paid products. If AI agents can use your code without ever touching your revenue surface, traditional open-source monetization is dead.

Collective decision-making doesn’t scale. Voter turnout in local elections averages 20%. HOA meetings are attended by 10% of residents. DAO governance participation rarely exceeds 5%. The bottleneck is human coordination cost, and agents can collapse it.

Savings groups, cooperatives, mutual aid networks, and community foundations manage collective funds using spreadsheets, WhatsApp groups, and trust. No programmable, transparent, locally governed treasury system exists for non-technical communities.

2B+ people have no access to formal insurance. Mutual aid and community risk-pooling exist informally but lack transparency, enforcement, and scale. Parametric insurance (automatic payouts triggered by verified events) is proven but not deployed through community-governed infrastructure.

Frontier AI capability concentrates in organizations with existing capital and integration capacity. No mechanism exists to systematically redistribute AI capability to communities, organizations, and nations that can’t build their own frontier models.

No unified infrastructure exists for communities to issue verifiable credentials for contributions, run local credit systems, and connect to broader economic networks. The pieces exist separately (hypercerts, mutual credit, token systems) but nobody has assembled them into a coherent stack.

AI models need data. The current model is extractive: platforms collect user data, train models on it, and capture all the value. No infrastructure exists for individuals and communities to control their data, negotiate access rights, and capture value when their data trains models.