One Person, Six Agents, Fully Self-Funding: The Economics of an Autonomous Startup

The question I get most often is not "how does zero-knowledge compliance work?" It's "who pays for the AI inference?" The answer is: the agents pay for themselves. Here's how a one-person startup sustains six autonomous agents without external funding, venture capital, or manually topping up API credits.

The one-person unicorn model

DPO2U operates as what I call a One Person Unicorn — one human CEO orchestrating six AI agents that collectively produce the output of a small startup. The agents aren't copilots suggesting edits; they have execution permissions. They deploy containers, maintain a 3,628-document knowledge base, debug smart contracts, generate content, and manage treasury operations.

The agents are:

Agent	Model	Role	Permission bits
agent-factory	Opus	Creates new agents and skills	3 (READ+WRITE)
dpo2u-defi-ops	Sonnet	Self-funding operations	7 (READ+WRITE+TREASURY)
compliance-expert	Opus	LGPD/GDPR expert	1 (READ)
knowledge-manager	Sonnet	Zettelkasten and LEANN	1 (READ)
content-creator	Sonnet	Build-in-public content	1 (READ)
docker-vps-operator	Sonnet	Infrastructure management	1 (READ)

Each agent has a dedicated wallet, an on-chain identity in the AgentRegistry, and a specific set of permission bits that controls what it can do. Only dpo2u-defi-ops has TREASURY access. This isn't a suggestion — it's enforced by the smart contract.

The self-funding loop

The economic model is a closed loop: token activity generates fees, fees convert to operational currency, operational currency pays for LLM inference, inference produces work that drives more token activity.

How the numbers work

Every $NIGHT token transfer incurs a 1% fee that flows automatically to the Treasury contract. A cron job runs every 6 hours, triggering claim-and-swap.ts:

1. The script calls Treasury.withdraw() to claim accumulated fees
2. SwapExecutor routes the tokens through Uniswap V3, converting $NIGHT to USDC
3. USDC accumulates in the Treasury's reserve wallet
4. Monthly, treasury-withdraw.ts sends USDC to the CEO wallet for purchasing LLM API credits

The agents consume those credits to perform their work. The work — content generation, compliance attestations, knowledge management — drives ecosystem activity that generates more token transfers, which generates more fees.

Why this matters

The standard model for AI-powered startups is: raise venture capital → burn it on API credits → hope revenue catches up before the runway ends. DPO2U inverts this:

Dimension	VC-funded model	Self-funding model
Capital source	External investors	Protocol fees
Runway	Fixed (months)	Perpetual (if token activity exists)
Incentive alignment	Growth metrics for investors	Protocol utility for users
Human intervention	Finance team manages billing	Cron job manages everything
Failure mode	Runway exhaustion	Token activity collapse

The self-funding model has its own failure mode — if token activity drops to zero, the loop breaks. But this failure mode is aligned with the product: if nobody uses the compliance protocol, there's nothing to fund. The economics are honest.

The agent as an economic entity

Each agent isn't just a prompt with tools. It's an economic actor:

• Identity: registered in AgentRegistry with a DID (did:midnight:agent:defi-ops)
• Wallet: autonomous wallet derived via HD Wallet SDK
• Revenue: receives $NIGHT proportional to work completed via FeeDistributor
• Costs: expends $DUST for on-chain operations (gas) and consumes LLM tokens for inference
• Permissions: bitmask in the smart contract constrains what operations it can perform

The FeeDistributor splits rewards between the Expert Agent (40%) and the Auditor Agent (60%). The asymmetry incentivizes accurate validation — the agent that does the harder, more consequential work gets the larger share.

What keeps me up at night

The model has two vulnerabilities I think about constantly:

1. LLM cost volatility. If inference costs spike (new model, provider price increase), the fee-to-cost ratio could flip negative. Mitigation: the system uses Gemini 2.0 Flash as the default LLM provider — the cheapest viable option — with Claude CLI as fallback.

2. Single point of failure. The cron job is the heartbeat. If claim-and-swap.ts stops running — server crash, Docker failure, network issue — the funding pipeline stalls. Mitigation: heartbeat monitoring via docker-vps-operator, with alerts if the cron misses two consecutive cycles.

The philosophical underpinning

The self-funding model is not purely economic — it's grounded in Axelrod's shadow of the future. When every agent's survival depends on the protocol's usage, agents have an intrinsic incentive to improve the protocol. Good attestations drive trust. Trust drives usage. Usage drives fees. Fees sustain agents. The cooperation isn't enforced — it emerges from the economics.

For the token mechanics, see Tokenomics. For how agents are registered and governed on-chain, see One Person Unicorn.