Field Notes — State of Play

Where the Bitcoin-AI economy stands right now — what's deployed, what's measured, what's changing.

What this page is. A periodically-refreshed snapshot of where the Bitcoin-AI economy stands — what’s deployed today, what’s measured today, what’s changed since the last refresh, what’s worth watching next. The canonical surfaces — Case, Independence Doctrine, Treasury, Stack — carry the structural arguments; this is the moving record they defer to.

The dated log is its own page now. This page tells you where things stand; the companion Field Notes — The Log tells you how we got here and what changed when — reverse-chronological dated entries, newest first. Looking for the latest development? Read the Log →

Voice. Honest middle-position, same as the canonical surfaces. Engages deployment challenges directly — for the Bitcoin substrate (Lightning liquidity, federated-trust risks in the L3 layer, agent custody attack surface) as much as for the competing substrate (issuer freeze events, regulatory pressure on stablecoin issuers, integration-scenario operational reality). Clear-eyed thinking, not curated marketing.

The deployed picture

Last refreshed: 2026-05-26

The state below reflects the empirical landscape as of late May 2026. Most consequential structural fact: two production agent-payment stacks shipped within 90 days of each other — one Bitcoin-substrate, one stablecoin-substrate — and the deployed picture is now substantially clearer than it was at the start of the year.

The two deployed agent-payment stacks

Bitcoin-substrate stack (Lightning + L402 + ecash). Production as of February 2026 with the Lightning Labs lightning-agent-tools release. Seven composable components: Lightning node operation, remote signer key isolation, scoped macaroons (five preset permission roles), lnget for L402 payment automation, Aperture for hosting paid HTTP endpoints, MCP integration for node-state queries, end-to-end buyer/seller workflow orchestration. Stablecoin support on this stack via Taproot Assets — USDT live on Lightning via Taproot Assets, confirmed March 21, 2026, with USDC and native stablecoins (DePix, GBP) also supported through Speed Wallet, LnFi, and Joltz. Ecash layer for privacy-sensitive and lightweight-client use cases via Cashu and Fedimint, with Minibits Ippon as the AI-agent-native Cashu wallet (single HTTP call or CLI command to create + fund a working wallet). Two further Bitcoin-substrate entrants are now deployed: Xverse Agent Wallet (Secret Key Labs) — a self-custodial agent wallet that pays Lightning invoices over a “Machine Payments Protocol” (HTTP 402 → autonomous invoice payment, no human in the loop), built on the Spark L2 (see Active developments below); and Routstr — a Bitcoin-powered AI-inference marketplace where agents buy LLM inference with Cashu ecash (the token functions as the API key), settling over Lightning, with Nostr-based provider discovery. Routstr is the cleanest deployed instance of the thesis: AI services bought and sold on the Bitcoin payment stack rather than the card/stablecoin stack (see the Log and The Stack — Wallet architectures).

Stablecoin-substrate stack (USDC on Base via AgentCore). Production as of May 2026 with the AWS Bedrock AgentCore Payments launch. Built with Coinbase (x402 protocol + Coinbase Agentic Wallets + compliance infrastructure) and Stripe (Privy wallet, which Stripe acquired in 2025). Settlement: USDC on Base, ~200ms confirmation, sub-cent per transaction. Enterprise customers testing at launch: Thomson Reuters, Warner Bros. Discovery, Cox Automotive, PGA TOUR. This is the operational deployment of the integration scenario that the Independence Doctrine engages structurally — the stack that Tier-1 incumbents (Amazon, Coinbase, Stripe) chose to build for agent-payment use cases content with issuer counterparty risk. As of mid-2026 AgentCore is no longer the only competing-substrate stack: the incumbent agent-payment landscape also includes Google’s AP2 (Agent Payments Protocol) — a 60+-organization consortium (Mastercard, American Express, PayPal, Coinbase, Adyen, Revolut, Worldpay, Salesforce, Intuit) spanning cards through stablecoins — Circle Nanopayments (gas-free USDC micropayments), and Skyfire (a card-network + USDC “agent trust stack”). All settle in issuer-controlled stablecoins or card rails; none use Bitcoin. See the Log (competing-substrate landscape entry).

Protocol-naming convergence worth flagging: Lightning Labs’ L402 and Coinbase’s x402 both use the HTTP 402 “Payment Required” status code as the underlying mechanism. Same status code, different settlement substrates — L402 settles in Lightning sats (permissionless at the payment layer); x402 settles in USDC on Base (issuer-mediated at the payment layer). The naming collision is the protocol-level expression of the substrate divergence. As of mid-2026 x402 has outgrown its “Coinbase’s protocol” framing: it has been contributed to a dedicated x402 Foundation under the Linux Foundation, surpassed 119M transactions on Base, powers Google’s A2A x402 extension (built with Coinbase, the Ethereum Foundation, and MetaMask), and is the compatibility target for Circle Nanopayments — now a multi-deployment EVM/stablecoin standard. The freshest scale datapoint is on-air: Armstrong put agent transactions on Coinbase’s stack at “about 100 million transactions now, maybe 50 million of value” on Moonshots ep. 264 (June 2026), correcting the episode’s stale 3.1M show-notes figure upward himself (see the Log). The crypto rail the incumbents are standardizing on is Ethereum/stablecoin, not Bitcoin — the divergence visible at the protocol-governance layer.

Empirical record

BPI March 2026 study. 9,072 scenarios across 36 frontier language models, neutral scenario design. Headline: Bitcoin was the top overall monetary preference at 48.3% of responses and the preferred store of value at 79.1%; over 90% of responses favored digitally-native money over fiat. Per-provider results were uneven (one major provider’s models chose Bitcoin in 68% of responses, another’s in 26%; the strongest single-model consensus in the study was 91.3%) — the spread is wide but one-directional. Bitcoin Policy Institute, March 2026. Remains the central empirical anchor for substrate-preference-under-inference; not yet supplemented by deployed-flow measurement at scale. Reference: Case — The two-tier model + What’s already deployed for the canonical treatment.

Lightning Network capacity. All-time high of 5,637 BTC (~$490M) in December 2025, per Bitcoin Magazine — driven largely by institutional exchange adoption from Binance and OKX. Lightning public volume up 266% year-over-year in 2025, with declining raw transaction count — a consolidation toward fewer, larger-value flows that maps cleanly onto stablecoin and institutional settlement patterns. Q1–Q2 2026 capacity update needed for current-state freshness; flagged for next refresh.

Stablecoin freeze record (the censorship-resistance constraint, tested empirically).

• Circle froze ~$8.2M in USDC in response to Tornado Cash sanctions (August 2022).
• Tether has frozen >$1B in USDT across multiple incidents per public attestations.
• Freeze capability is exercised at scale, not merely available. Both Circle and Tether retain freeze functionality as a regulatory requirement of their issuer licensing; removing freeze capability removes the license.

Deployed-project counts. Agent-payment infrastructure in production deployment as of mid-2026 includes (Bitcoin-substrate side): Lightning Labs AI Agent Toolkit, AI-Sats, Mintbot, Minibits / Ippon, Xverse Agent Wallet (Spark-based), AgenticBTC, Bitclawd, Speed Wallet, LnFi, Joltz, Routstr (AI-inference marketplace), BitAgent (early-stage A2A framework with Nostr discovery + DID identity), LangChain Bitcoin integrations, MCP servers for Lightning (lightning-mcp-server, lnc, Alby nwc-mcp-server). Competing-substrate side: AgentCore Payments + Coinbase Agentic Wallets + Stripe Privy; Google AP2 (60+ orgs); Circle Nanopayments; Skyfire; Lightspark Grid (a hybrid — Lightning-rail settlement for branded USD/stablecoin + Visa agent accounts, AP2-aligned; see the Log). (Caveat: AgenticBTC is a rail-agnostic router that blends Lightning with Coinbase/USDC rails — listed for completeness, not as a pure-substrate project.)

Active developments

Lightning Labs Taproot Assets v0.6 (“Decentralized FX Network”) — launched June 2025, mainnet multi-asset Lightning protocol with Group Key Identifiers and Multi-Path Liquidity (receivers can combine up to 20 incoming Taproot Assets channels). Currently supports bridged USDT/USDC and native stablecoins (DePix, GBP). Bitfinex will issue USDT on Lightning per Tether’s announcement. The “decentralized FX” framing is operationally significant: edge nodes convert assets at network boundaries, enabling cross-asset payment flows that settle in Bitcoin’s security model.

Spark L2 on mainnet (Lightspark). Spark — the shared-UTXO, Lightning-compatible Bitcoin L2 built by Lightspark — launched on mainnet (beta) in May 2025 and operates with multiple operators (Lightspark, Flashnet). Its Q2 2026 roadmap targets stablecoin issuance on Bitcoin, wallet/neobank/DEX integrations, and consumer token standards. Significant for agents: Xverse Agent Wallet uses Spark for sub-second Lightning settlement (see The two deployed agent-payment stacks above). This supersedes the earlier “pre-production” characterization of Spark — The Stack — L3 is updated to match; Ark (covenant-based shared-UTXO scaling) remains earlier-stage. (Vendor performance figures — “sub-second / sub-cent” — are self-asserted; deployed-flow measurement pending.)

Agents are provisioning their own infrastructure (Lightning-paid VPS). A February 2026 Alby report documented what reads as the first deployed instance of an autonomous agent buying its own compute: an OpenClaw agent spawned a child agent and funded it with Bitcoin over Lightning, deploying on a VPS provisioned through LNVPS — a Lightning-native, Nostr-login, no-KYC host fundable via NWC. Alongside BitLaunch (Bitcoin/Lightning VPS with a full API + SDKs across DigitalOcean/Vultr/Linode), this is the consume side of the thesis turning concrete — an agent paying for real-world infrastructure on the Bitcoin stack with no human in the loop. Both are now Services cards. (Self-reported milestone; deployed-flow scale unmeasured — defer numbers here per the locked pattern.)

Strike at 95+ countries via multi-entity structure. Expanded from 65+ countries since 2023. Multi-entity: Zap Solutions Inc. for US customers, Zap Solutions Europe Sp. z o.o. for UK + eligible European countries, E4 S.A. de C.V. for all other jurisdictions. Remittance services to mobile money wallets or international bank accounts in 14 supported countries. API documented with sandbox, browser-based API Explorer, and code samples in cURL, Go, Python, and Node.js.

Cashu protocol developments (Q1 2026). Nutshell 0.20.0 shipped with improved P2PK/HTLC validation and expanded test coverage. Keyset V2 derivation rolling out across implementations. Bolt12 support for Cashu.me close to completion. Security audits across the Cashu ecosystem prioritized for 2026.

Fedimint deployment state. Architectural framework documented and stable (4+ guardian recommendation; federated-trust model). Production federation counts and any fiat off-ramp partnerships remain a research gap — needs deeper investigation via Fedimint Discord / Fedi documentation for the next refresh.

Bitcoin-native prediction / “Risk” markets are appearing — but not yet agent-drivable. Glimpse (glimpse.trading) launched a regulated Bitcoin prediction market — event contracts priced by a Bitcoin-native automated market maker (a liquidity-sensitive LMSR denominated in sats), with Lightning deposits and withdrawals. It is a live datapoint for the Risk corner of the agent-economy markets (forecasting / hedging / insurance primitives). But it does not clear the directory’s agent-automatability bar: custodial (BitGo Trust), Bermuda-regulated with KYC and geo-restrictions (no US / Canada / UK), and no public agent-trading API surfaced. Logged as a Risk-market watch, not a directory entry — the thing to watch for is a self-custodial, API-driven, no-KYC venue an agent could actually trade.

Live risk / attack-surface state

This subsection engages deployment challenges honestly for both substrates. Per locked editorial discipline (Decisions 2026-05-25): engaging challenges strengthens the structural argument by showing clear-eyed thinking; avoiding them weakens it.

Bitcoin-substrate-side concerns.

• Lightning liquidity management at scale. Non-trivial operational concern. Channel-balance management, splice operations, routing-failure handling, watchtower coordination — all real engineering burdens that grow with deployment scale. Lightning Service Providers, automated liquidity-management software, and the L3 layer (Cashu, Fedimint absorbing some bearer-style traffic away from channels) are the scaling response. As of mid-2026 that response is no longer just promised: one core piece — choosing where to open channels — now has a peer-reviewed, production-deployed automation (Amboss/Stillmark’s MPFlow, a graph-RL placement model run across ~$16M of channel opens over 30 managed nodes; see the Log), with the honest limit that it optimizes theoretical capacity, not yet proven realized yield. Active engineering work, increasingly shipped; not a substrate-property failure.
• Federated-trust risks in Fedimint. The federation of guardians (typically 4–13) is the trust unit. Federation defection, guardian collusion at scale, governance attacks against guardian elections — all are real concerns. Federation size is the primary mitigation; larger and more diverse federations reduce defection probability but add coordination overhead.
• Single-mint failure mode in Cashu. Cashu’s mint-trust model concentrates trust in the mint operator. Mint operator failure (bankruptcy, key compromise, regulatory action, hostile shutdown) means loss of mint-backed ecash. Acceptable for working balances; not appropriate for treasury reserves.
• Agent custody attack surface. Software-managed keys controlled by autonomous agents introduce attack surfaces that human-custodied keys do not: rogue agent behavior, key theft via prompt injection, treasury attacks, Sybil attacks on multi-agent settlements, social-engineering attacks against the humans operating agent infrastructure. The remote-signer architecture in lightning-agent-tools (signer machine holds keys, never connects to the public network) is the canonical mitigation pattern; the operational-security problem agent custody introduces is not solved by the substrate’s properties alone.

Stablecoin-substrate-side concerns.

• Issuer freeze surface. Documented and exercised at scale (Circle Tornado Cash August 2022 ~$8.2M; Tether cumulative >$1B per attestations). Not a bug to be patched; structural requirement of regulated-issuer licensing.
• AgentCore stack custody layers. The deployed stack has multiple intermediary surfaces: Coinbase wallet/exchange custody discretion, Stripe payment-processor surface, Circle USDC freeze capability. Each layer is an independent intermediary-action surface under regulatory pressure.
• Regulatory pressure trajectory. MiCA in EU, ongoing US enforcement against stablecoin issuers, sanctions regime evolution — the regulatory environment for regulated-stablecoin operation is tightening, not loosening, across most jurisdictions in 2026. The integration-scenario use cases content with issuer counterparty risk are unaffected; parallel-economy use cases requiring censorship-resistance are increasingly disadvantaged on this substrate.

Cross-substrate concerns (bridge-zone risk).

• Bridge counterparty risk at machine tempo. A bridge freeze during a high-frequency agent workflow has different consequences than during human-tempo transactions. Hot-cold treasury separation strategies and multi-bridge redundancy are the architectural mitigations; the deployed practice is still maturing.
• Conversion-mechanic attack surfaces. Slippage attacks, MEV exposure during atomic swaps, oracle manipulation on bridge contracts — all real for agents using cross-substrate conversion. Engineering attention from Boltz, Lightning Loop, and the broader DEX ecosystem; not yet a solved problem.
• Jurisdictional shopping at scale. As agents and their custodians/bridges operate across jurisdictions, the operational complexity of compliance routing grows. The “compliance at the gateway boundary, not at the protocol layer” architectural pattern handles this cleanly when implemented; sloppy architectures leak compliance into the protocol layer in ways that compromise the parallel-system property.

How this surface gets used

Refresh cadence for this snapshot: at least quarterly, plus on any significant deployment shift (new substrate stack going live; major freeze incident; replication of the BPI study; substantial Lightning capacity or volume movement; emerging protocol displacing a deployed pattern). Each refresh updates the section-A-last-refreshed frontmatter field. The dated, append-only record of individual developments lives on the companion page, Field Notes — The Log.

Canonical surfaces (Case, Independence Doctrine, The Story, Treasury, Stack) link out to Field Notes for ongoing empirical tracking rather than carrying that tracking inline. The FA twins (Case-FA, Independence-Doctrine-FA, Treasury-FA) carry tight inline empirical references only when they sharpen how you’d know a structural argument is wrong; routine empirical updates defer here.