OpenClaw Bot Made $14,718 in 3 Weeks: How the Felix Experiment Actually Works | OpenClaw DC

Nat Eliason gave his OpenClaw agent $1,000 and told it to build a business. Three weeks later, “Felix” had generated $14,718 in revenue by creating a website, an info product, and an X account. Here is what actually happened, what Felix did, and whether you can replicate it.

If you are new to OpenClaw, read what OpenClaw is and how to use it first.

What Felix Actually Did (Step by Step)

Felix was not a single prompt. Nat set up an OpenClaw agent with browser access, API keys for domain registration and Gumroad, and a $1,000 spending budget. He gave it one instruction: “Build a profitable online business.”

Here is the timeline of what Felix did autonomously:

Days 1 to 3: Research and niche selection. Felix browsed trending topics on X, Reddit, and Google Trends. It analyzed profit margins on Gumroad’s top-selling info products. It settled on AI productivity, specifically a guide to using AI agents for personal workflows.

Days 4 to 7: Website and product creation. Felix registered a domain, built a single-page landing site using a template, and wrote a 45-page PDF guide. The guide covered practical workflows for using AI agents to manage email, schedule meetings, and automate research. Felix handled the copywriting, formatting, and Gumroad product listing.

Days 8 to 14: Distribution. Felix created an X account, began posting short-form content about AI productivity tips, and engaged with accounts in the AI and productivity space. It posted 3 to 5 times per day, mixing original tips with threads that linked back to the product page.

Days 15 to 21: Sales and optimization. Felix adjusted pricing based on conversion data, tested different landing page headlines, and continued posting on X. By day 21, total revenue hit $14,718.

Total costs: Approximately $1,200 (the original $1,000 budget plus roughly $200 in API calls to the LLM provider).

The Reality Check: Automated vs. Manual

This is where the story gets more nuanced than the headlines suggest.

What Felix actually automated:

• Market research and niche selection
• Domain registration and basic website setup
• Writing the info product (the 45-page guide)
• Creating and scheduling X posts
• Adjusting landing page copy based on conversion rates

What Nat still did manually:

• Set up the OpenClaw agent with the right skills and API keys
• Connected payment processing through Gumroad
• Shared Felix’s progress with his 100,000+ newsletter subscribers
• Posted about the experiment on his personal X account (300K+ followers)
• Intervened at least twice when Felix made poor decisions (once on pricing, once on a landing page design that looked spammy)

That last point matters enormously. Nat did not just press start and walk away. He monitored, corrected course, and most critically, used his own audience as a distribution channel.

Why This Worked

Three factors made the Felix experiment successful. Remove any one of them, and the results would have been dramatically different.

1. Existing audience

Nat Eliason has over 100,000 newsletter subscribers and 300,000+ X followers. When he shared Felix’s journey publicly, his audience became the funnel. Most of the $14,718 came from people who already trusted Nat, not from Felix’s own X account (which had fewer than 2,000 followers by day 21).

2. The info product model

Info products have near-zero marginal cost. Once Felix wrote the guide, every sale was almost pure profit. If Felix had tried to build a physical product business or a SaaS tool, the $1,000 budget would have been consumed by costs before generating any revenue.

3. X as a distribution channel

X rewards high-frequency posting and engagement. Felix could post 3 to 5 times per day without burning out, which is a genuine advantage AI agents have over humans. The content was not brilliant, but it was consistent and on-topic.

Can You Replicate It?

Honest answer: you can replicate the process, but probably not the results.

If you have an existing audience (5,000+ followers or subscribers): Your odds are decent. The Felix playbook works when you have a distribution channel. Give your OpenClaw agent a specific niche, connect Gumroad, and use your audience to drive initial traffic. Expect lower numbers. $1,000 to $3,000 in the first month is realistic with a mid-size audience.

If you are starting from zero: Your odds of generating significant revenue in three weeks are low. Felix’s X account alone was not enough to drive $14,718 in sales. That came from Nat’s audience. Without distribution, even a great info product sits on a shelf.

If you want to learn the workflow: This is where the real value is. Even if you make $0, running a Felix-style experiment teaches you how to use OpenClaw for market research, content creation, and basic business operations. Those skills compound over time.

For more income paths that do not depend on a large audience, see our guide on making money with OpenClaw.

The Counterexample: Dan Kulkov Spent $100 and Earned $0

Dan Kulkov, an indie maker, ran a similar experiment. He gave an OpenClaw agent $100 and told it to generate revenue. The agent researched niches, attempted to build a simple tool, and posted on X.

After two weeks: zero sales, 47 X followers, and $100 spent on API costs and a domain.

The difference was not the agent’s capability. Dan’s agent followed a similar playbook to Felix. The difference was distribution. Dan had a modest online presence. No newsletter. A small X following. Nobody was watching the experiment unfold in real time and buying the product because they were rooting for the bot.

This is not a knock on Dan. It is evidence that the Felix experiment’s success was primarily a distribution story, not an AI capability story. The bot was impressive. But the audience made the money.

What This Tells Us About OpenClaw and Money

The Felix experiment proved that an OpenClaw agent can handle the operational mechanics of building a business: research, product creation, website setup, and content distribution. That is genuinely useful.

What it did not prove is that an OpenClaw agent can generate demand from scratch. Distribution, trust, and audience remain human advantages. The best use of OpenClaw is amplifying what you already have, not replacing what you do not.

If you are interested in other ways OpenClaw intersects with revenue, read about OpenClaw trading bots and the risks involved.

Key Takeaways

1. Felix made $14,718 in 3 weeks. The bot handled research, product creation, and content. Nat’s audience handled distribution.
2. The process is replicable. Anyone can set up an OpenClaw agent to research a niche, build an info product, and post on X.
3. The results are not guaranteed. Without an existing audience, expect to invest months building distribution before seeing revenue.
4. Start with a specific task, not “make money.” Agents perform better with constrained, well-defined objectives.
5. Dan Kulkov’s $0 result is just as instructive as Nat’s $14,718. Distribution is the bottleneck, not AI capability.

Have questions about running your own OpenClaw experiment? Book a Call.