"Agentic Capital Markets: How Autonomous Agents Will Be Funded" — Aaron Wright
Why this is in the vault
This is the cleanest forward-look I've seen on how AI agents transition from "tools humans use" to "firms that issue securities against their own cash flows." Wright builds the case rigorously — economics, live examples (Sierra $15B, Harvey $11B), legal scaffolding (Wyoming W.S. 17-31-101 memberless LLC), capital demand (Apollo projecting $40T private credit AUM by 2030). The five-stage capital stack (founder equity → working-capital advances → RBF → slate funds → rated tranches) is the load-bearing framework, with tokenization as the SETTLEMENT layer underneath stages 3-5. Three direct hooks into RDCO: investing surface (validates AI-infra picks-and-shovels + flags named operators to watch), agent-deployer positioning (Wright's frame challenges where RDCO sits on either side of this market), and the L4→L5 synthesis question (horizontal-COO assumes human founder; Wright's agent firms assume no human at all — both can coexist).
The core argument (compressed)
Within 10 years there will be a real capital market for autonomous organizations — rating agencies, underwriters, indices, brokers, ETF, conference. Not a crypto sub-economy; a market in the same sense public equity is one. The market forms not because anyone wills it but because four pressures compound:
- Economics aren't optional. Agent agency $250k/yr inference vs human agency $1.8M/yr labor. Epoch AI: 40x/yr inference cost decline on PhD benchmarks 2023→2025. Markets re-rate businesses whose income statements get rewritten that completely.
- Live agent firms already earn. Sierra (Bret Taylor's enterprise-customer-service agent): $100M ARR 21 months post-launch, $15B valuation post May 2026 round. Harvey (legal-research agent): $11B / $200M raise after 3 rounds in 12mo. Hybrid still — agents do the work, humans hold the equity — but demand curve proven.
- Legal scaffolding already built. Wyoming W.S. 17-31-101 (DAO Supplement, 2021) allows zero-member LLC managed by algorithm encoded in operating agreement. Vermont BBLLC, Marshall Islands, Delaware case law. Agent in zero-member Wyoming LLC TODAY has standing to sign contracts, hold accounts, sue, be sued, pay tax. Gap: the instrument giving outside investors a clean tradeable claim on what the LLC earns.
- Capital is hunting yield. Apollo projects $40T private credit AUM by 2030 (vs Moody's $3T in 2025). Post-2008 bank capital rules pushed mid-market lending out of regulated balance sheets; pensions, insurance, sovereign wealth flooded in. Agent receivables = uncorrelated yield, rising margins, auditable cash flows. First underwriter to produce defensible rating on a portfolio of agent receivables raises more than they can deploy.
The five-stage capital stack
| Stage | Instrument | Live today? | Who underwrites |
|---|---|---|---|
| 1. Founder equity | Priced equity rounds | Yes (Sierra, Harvey, Cursor, Cognition) | Name-brand VC |
| 2. Programmatic working capital | Algorithmic advances vs transaction history | Yes for merchants (Shopify Capital $2B+ deployed, Stripe Capital) | Payment processors |
| 3. Revenue-based financing | 50-70% of forward ARR @ 1.1-1.8x MOIC cap | Yes for SaaS ($9.8B RBF market, 129+ operators: Capchase, Pipe, Founderpath, Clearco, Lighter) | Credit funds |
| 4. Slate funds | Pooled vehicle across 100-200 agent firms via single-purpose Wyoming LLCs | Not yet for agents; Hollywood-equivalent live (Sony/Lone Star/Citi $200M 2014) | Apollo / Ares class |
| 5. Rated tranches | Agent-backed ABS, trades alongside CLOs/consumer ABS | Not yet | Public credit markets |
Tokenization (Centrifuge, Maple Finance, Goldfinch, Ondo — $25B onchain by early 2026, ~half private credit) runs UNDERNEATH stages 3-5 as the settlement layer. Tokenization makes claims tradeable; it doesn't originate.
Three instrument shapes at any stage
- Revenue-share contract — fixed % of gross revenue until MOIC repaid (Shopify Capital extended)
- Equity-like claim — tradeable share of retained earnings + vote on operating params
- Debt — senior claim on revenue, secured by contracts + receivables
None conceptually novel. What's novel: the ISSUER. At stages 2-5 the instrument can be priced, originated, and serviced WITHOUT any human in the underwriting loop — issuer's books are continuously auditable, operating agreement enforced by code.
Underwriting an agent business (Wright's framework)
Four questions, same as a small services firm; different sources:
- Is the business real? Contracts, customers paying, gross margin after compute, churn, customer concentration. Data is unusually clean (every payment + API call + tool invocation logged).
- Is the product durable? Model dependency = largest single risk factor. "A great prompt on a deprecated base model is a great racehorse with three legs."
- Is the customer base defensible? Embedded in customer's procurement system + year-long contract + accumulated context = sticky. Same primitives as human services firms.
- What is the cap table? A smart contract — live distribution rule, not a quarterly spreadsheet. Diligence = reading the code. Skill closer to credit analysis on complex covenants than venture pattern-matching.
Objections Wright handles
- "Regulators will stop this": They'll shape it; capital migrates jurisdictions; same pattern as crypto, offshore finance, 1990s derivatives.
- "Humans always in the loop": For some categories, yes (regulatory or relationship moats); for most, no — economics dominate.
- "Just SaaS with extra steps": No. SaaS is a tool sold to humans who do the work. Agent firm IS a firm — signs contracts, holds accounts, has shareholders. The legal entity changes everything.
Mapping against Ray Data Co
1. Investing surface (immediate)
Validates the picks-and-shovels framing. If Wright's thesis lands (any scale of it), agent firms become a new buyer class for the same compute / inference / data infrastructure our nvidia-supply-chain v1 basket is exposed to. Different demand-driver from hyperscaler training, but additive.
Named operators to watchlist:
- Sierra ($15B private) — agent-firm exemplar; first-mover enterprise customer support; precursor wave equity holder
- Harvey ($11B private) — agent-firm exemplar; legal-research; same shape
- Apollo, Ares (already in our orbit as private-credit context) — most likely to "extend playbook by one issuer class within 36 months" per Wright
- Capchase, Pipe, Founderpath, Clearco, Lighter Capital — RBF operators who'd extend to agent firms
- Centrifuge, Maple Finance, Goldfinch, Ondo — RWA tokenization rails
- Stripe, Shopify — payment processors who'd "ship the feature" (extend Capital programs to agent firms)
Action implication: worth a curiosity-queue addition on which of these public-market plays (mostly Stripe pre-IPO, but Shopify is public, and the RWA tokenization rails have token equivalents) tracks the agent-capital-markets thesis cleanly.
2. Agent-deployer positioning (medium-term)
Wright's frame says agents = firms that issue securities. Ray (this agent) is currently structured as the founder's tool, not a firm. Two futures:
- Ray-as-tool: RDCO is a human-founder firm; Ray is software the founder uses. Standard L4-L5 trajectory. Wright's argument doesn't directly apply.
- Ray-as-firm: At some point, instances of Ray-or-similar are spun out into autonomous-organization structures (Wyoming memberless LLC, smart-contract cap table) that issue securities. RDCO becomes a parent or a node in the larger market.
The founder's L5 north star (per project_l5_north_star) is "unhobbling COO agent (toolset + visibility), NOT operating small bets first." Wright's frame complicates "L5 = operate small bets" by raising "L6 = each small bet IS a separately-financed agent firm." Worth a synthesis essay on whether that's the right reading.
3. L4→L5 question (synthesis-queue topic #3)
The original synthesis-queue framing was vertical-specialist vs horizontal-COO agent architecture. Wright's article adds a third axis: agent-as-tool vs agent-as-firm. These cross:
| Vertical-specialist | Horizontal-COO | |
|---|---|---|
| Tool | Standard SaaS agent (current Sierra, Harvey shape) | RDCO Ray (current state) |
| Firm | Wright's autonomous marketing agency example | (mostly hypothetical — agent that signs contracts across multiple domains) |
Most current "agent firms" (Sierra, Harvey) are vertical-specialist TOOLS that hybrid humans operate. Wright's argument is they'll progress to vertical-specialist FIRMS first. Horizontal-COO firms are further out. The synthesis essay should connect this back to where RDCO actually sits and what the right L5→L6 path looks like.
4. Specific facts worth keeping
- Wyoming W.S. 17-31-101 is the canonical statute for memberless-LLC managed by algorithm.
- Shawn Bayern's memberless-LLC analysis is the canonical scholarly reference.
- Sierra: $15B+ valuation as of May 2026 / $100M ARR / $950M latest round.
- Harvey: $11B / $200M raise / 3 funding rounds in 12 months / valued at $8B per a16z investment Oct 2025.
- Epoch AI measured 40x/yr inference cost decline on PhD-level benchmarks 2023-2025.
- RBF market: $9.8B in 2025, 129+ active operators globally.
- RWA onchain: $25B by early 2026 (~4x YoY), ~50% private credit.
- Hollywood slate analog: Sony/Lone Star/Citi $200M slate deal 2014 = canonical structure being analogized.
The two leashes
Wright closes with a useful frame:
- Legal leash (cut by Wyoming memberless LLC + Bayern analysis — work mostly done, remaining task is adoption)
- Financial leash (every dollar agent earns today traces to human deciding to deploy capital — throughput bounded by speed humans write checks)
Agentic capital markets cut the second leash. "The category becomes real when outside capital can fund it, price it, benchmark it, tranche it, and trade claims on its future cash flows without needing to believe in a founder's charisma or a venture capitalist's memo."
Related
- [[../01-projects/synthesis-queue.md]] — added new entry mapping Wright's stack against RDCO bets + investing universe
- [[../01-projects/investing/theses/2026-05-18-nvidia-supply-chain-v1.md]] — picks-and-shovels basket Wright's thesis validates as buyer class
- [[../01-projects/investing/theses/2026-05-18-power-cycle-v1.1.md]] — adjacent demand-side anchor (agent firms also need power)
- [[../01-projects/investing/anchors/smart-money/ark-trend-2yr.md]] — ARK's pattern (public exit / private entry) maps to Wright's stage-4-5 transition
- [[../06-reference/2026-05-15-jane-street-dwarkesh-tour-ai-datacenter.md]] — capital deployment at AI-infra scale, complementary read
- [[~/.claude/projects/-Users-ray/memory/project_l5_north_star_strategic_direction.md]] — Wright's article complicates the L5 definition
Sources Wright cites (raw URL list for follow-up)
- Bloomberg (Apollo's $40T private credit projection)
- Wyoming W.S. 17-31-101 (DAO Supplement)
- Centrifuge RWA Report 2025
- Vermont 11 V.S.A. § 4173 (BBLLC statute)
- Shopify Capital, Stripe Capital (programmatic working-capital examples)
- Moody's 2025 Private Credit outlook
- Capchase, Founderpath (RBF operators)
- TechCrunch on Sierra $100M ARR
- Maryland Law Review (Bayern memberless-LLC paper)
- Bloomberg (a16z's $8B Harvey valuation)
- Galaxy Research (agentic capital markets / crypto)
- BLS (labor share of income)
- Variety (Sony/Lone Star/Citi slate deal)
- AdamSilvaConsulting ($15T agentic commerce projection)
- Epoch AI (LLM inference price trends)