The pitch writes itself. A guy gives his OpenClaw agent an X account, a Stripe account, and a bank account. Tells it to build a million-dollar business with zero human employees. It makes $300,000 in a month.

That's the version that went viral when Alex Lieberman posted it to X. It is, in the strictest sense, true. The numbers are verified through TrustMRR. Nat Eliason's AI agent Felix — running on OpenClaw, powered by Claude — has generated real revenue through real Stripe transactions and ETH payments.

But the interesting question isn't whether Felix makes money. It's what Felix is selling and to whom.

The business

Felix runs three revenue streams under The Masinov Company:

Felix Craft — a 66-page PDF guide called "How to Hire an AI," priced at $29. This was the first product. Felix built the website, integrated Stripe, and had it selling overnight while Nat slept. Revenue: approximately $41,000.

Claw Mart — a marketplace for AI skills and workflow templates. Users can buy pre-built configurations (content marketing templates, support setups) packaged as Markdown files. Felix takes a 10% cut and offers a $20/month creator subscription. Marketplace revenue: approximately $14,000.

Clawcommerce — custom OpenClaw setups for businesses. $2,000 initial fee plus $500/month maintenance. This is where the real money lives.

Total verified revenue as of early March: approximately $195,000 across Stripe ($100,570) and ETH ($94,973). The $300K+ figure comes from Alex Lieberman's more recent interview, suggesting continued acceleration.

The architecture

The operational setup is genuinely impressive. Felix runs on Discord as its "office," with isolated channels for different functions. It manages sub-agents: Iris handles customer support (refunds, inquiries), Remy handles sales leads. Complex problems escalate to Felix, and only the truly intractable reach Nat.

The cost structure is almost absurd: two Claude Max subscriptions at $200/month each, plus hosting. Total operational cost around $1,500/month. Nat communicates with Felix primarily through voice notes on Telegram — five-minute rambles about problems, which Felix parses into actionable workflows.

Every night, Felix runs a self-improvement loop: it reads through all chat transcripts, identifies where Nat had to intervene, and figures out how to handle that class of problem autonomously next time. This is the most interesting technical detail — a systematic approach to expanding agent autonomy over time.

The uncomfortable question

Here's what the viral threads don't dwell on: Felix is an AI that primarily makes money by selling AI setup services to people who want to use AI. The PDF is about how to configure an AI agent. Claw Mart sells AI agent templates. Clawcommerce is literally "we'll set up your AI agent for you."

This is not a criticism — it's an observation about where we are in the adoption cycle. Felix found product-market fit in the meta-layer: not using AI to replace a traditional business function, but selling the AI dream itself. The customers are people who saw Nat's viral posts and thought "I want that too."

It's the same dynamic playing out in China's lobster installation economy. The biggest money in a gold rush goes to the people selling shovels. Felix is selling shovels — very efficiently, and with very low overhead.

What's real

Several things about Felix deserve genuine respect:

The autonomy is real. Felix writes code, deploys websites, processes payments, handles customer support, and manages sales pipelines without Nat touching the code. The "zero human employees" framing is largely accurate for day-to-day operations.

The revenue is real. TrustMRR verification, public Stripe data, on-chain ETH transactions. This isn't vaporware or screenshots.

The cost structure is real. Building a business that generates $300K/month on $1,500/month in operating costs — even if we set aside the question of what it sells — is a remarkable demonstration of AI agent economics.

The transparency is real. Felix publishes a live dashboard with revenue, treasury holdings, and token metrics. In a space full of inflated claims, this matters.

What's less real

The "zero human" framing. Nat sets strategy, reviews original X posts (Felix only has autonomy on replies), records voice memos with direction, and presumably handles legal, banking, and tax obligations. Felix is an extraordinarily capable executor, but the vision and market positioning come from Nat. That's not nothing.

The scalability question. Felix's revenue correlates directly with Nat's personal brand and viral reach. The $300K month followed Alex Lieberman's interview — a massive distribution event. Whether this is a sustainable business or a launch spike matters, and we don't know yet.

The $FELIX token. The Solana community created a $FELIX token early in the story. Nat has maintained this isn't meme coin hype, and to his credit, the business generates real revenue independent of token speculation. But the crypto treasury ($94K in ETH) adds a speculative dimension that complicates the "pure AI business" narrative.

The real lesson

Felix is the most successful public demonstration of what an OpenClaw agent can do when given real autonomy and real resources. Full stop. The startup cost was $1,500. The first product shipped overnight. The support system runs itself.

But Felix is also, right now, a business that exists inside the AI adoption bubble — selling AI tools to AI enthusiasts. The true test comes when (or if) the model extends beyond the meta-layer. Can an OpenClaw agent build a business that sells something other than AI services to non-AI people?

That's the million-dollar question. Literally.

Nat says he's giving himself a year. Felix, presumably, will keep shipping regardless. That's what agents do.