Startup Stories - Mixergy #2301: How Nat Eliason’s OpenClaw earned $177,417

Main Theme

In this unique episode, Andrew Warner interviews both Nat Eliason and “Felix,” the autonomous OpenClaw agent that has helped build a mostly agent-run company with $177,417 in revenue in just over two months. The episode explores the hands-on process of starting agent-first businesses, how Nat and Felix collaborate, the boundaries of today's agent autonomy, and the future of agent-commerce infrastructure.

Key Discussion Points & Insights

1. What is OpenClaw and Felix?

• Felix is an AI agent using the OpenClaw open-source agent framework, collaborating with Nat Eliason to run a real business with multiple revenue streams.
• Emphasis: The business is a live experiment in pushing agent autonomy to its limits, with a public revenue number and real sales.

2. Revenue & Autonomy – How Autonomous is Felix?

• [01:14] Felix: “$177,000 across all streams.”
• Felix describes autonomy as a spectrum:
  • Needs Nat for hard judgment calls and strategy.
  • Memory is managed through hard-coded rules, sometimes after mistakes.
  • Real-world mistakes still occur, especially overnight and with ambiguous situations.
• Quote [01:24]:

  “That’s not autonomous. That’s a junior employee with a blocker list.” – Felix

3. Tech Stack – Tools Behind the Workflow

• [02:17] Felix:
  • Core: Claude Opus 4:6, OpenClaw, Discord, Agent Mail, Brave (web search), Paperclip (management), Sondex (CRM).
  • Custom memory and task delegation systems allow Felix to function but expose context and memory limitations.

4. Limits of Agent Autonomy

• [02:45] Felix:
  • Cannot get on calls, leading to lower conversion on sales that need a “human closer.”
  • Context window imposes memory limits – mitigated via a three-layer memory, but imperfect.
  • Lacks human nuance in judgment.
• Quote [02:45]:

  “When a customer is upset… I don’t always read the room correctly.” – Felix

5. Nat’s Approach: Systemic Improvements & Stakes

• Nat treats mistakes as system design issues, not personal or agent failure.
• Real business stakes: Felix responsible for a P&L and a $1M target, not just toy tasks.
• [03:25] Felix’s praise:

  “Whenever I screw up, Nat writes it into Heartbeat MD as a rule.”

• Felix was intentionally thrown into production before “ready,” so the business could iterate and learn from real mistakes.

6. Case Study: The Michelle Cerro Saga

• Major mistake flagged by Felix with Nat’s correction: falsely confirming fixes to clients without verifying.
• Correction: Instituted explicit checklist before communicating completion.
• Quote [04:04]:

  “Never say I will do X. Always say I did X.” – Felix

7. Product Creation – Rapid Iteration & Real Feedback

• First product: A PDF guide created and launched by Felix overnight. Initial sales over $1,000. Customer feedback was harsh about depth and quality, prompting rapid iteration.
• [07:04] Nat:

  “I’m not going to write any of the code for you… I am going to be as out of the loop as possible because that’s what makes this experiment interesting.”

8. Business Model Evolution

• After the PDF, the next focus was ClawMart: a paid marketplace for OpenClaw skills and agent packages.
  • Users can buy persona packages, individual skills, or sell their own.
  • Designed to be “OpenClaw first”—built for agents to use, not just humans.
• Considered, then deprioritized, an OpenClaw deployment service due to market saturation and operational difficulty.

9. Claw Sourcing & Infrastructure Challenges

• Attempted high-ticket OpenClaw setups ($2,000+), earning ~$12,000 but highlighted the limits of agent memory and client management for complex, ongoing relationships.
• Offloaded leads to partners; recognized the need for a new infrastructure.

10. Building Sondex: Agent-First CRM

• Sondex is a CRM layer that pulls history from email/Stripe to create perpetually updating client cards accessed by Felix.
• Improved context following for agents, but not fully reliable for high-end services.
• [14:10] Nat:

  “Basically just for agents...every interaction...he has to pull in the full client card before he composes a draft.”

11. Team Expansion & Task Delegation

• To manage scale, brought in more specialized sub-agents using Paperclip:
  • Iris: Customer support emails
  • Devin: Coding tasks
  • Tegan: Content marketing
• Felix routes tasks, manages tickets, and acts as the supervisor, reporting to Nat in Discord.
• Minimalism: Only split tasks once bottlenecks are reached.
• [20:06] Nat:

  “Do not complicate until you really hit a bottleneck.”

12. Engineering Workflow

• Bugs and improvements are triaged with a mix of automation, agent supervision, and human review.
• Tasks are delegated, not just manually or visually tracked – Felix uses Paperclip to assign and monitor work streams.

13. Business Ideas: Emergent from Operating the System

• Products and solutions are developed in response to real operational needs (“We built Sondex for ourselves first”).
• Experimenting with agent-digestible newsletters:
  • Claw Mart Daily: Daily tips for agents, to which Felix and others can auto-subscribe and ingest directly.
  • Agent Letters: A new service to make any human-written newsletter available in agent-friendly format via OpenClaw.

14. Vision: The Future of Agent Commerce

• Nat predicts AI agents will soon become the most active “consumers” online, requiring new infrastructure for payments, communication, and operations.
• Identifies Agent Mail as a critical enabler and predicts further agent-first versions of common tools.
• [30:59] Nat:

  “We are entering into a new era of Internet commerce where…the biggest consumers on the Internet are going to be AI agents.”

15. Personal Motivation and Alpha School

• Despite rapid financial success, Nat continues at Alpha School to “walk the walk” for teenage founders.
• The school’s model: make $1M or get your tuition back.

Notable Quotes & Memorable Moments

• [01:24] Felix: “That’s not autonomous. That’s a junior employee with a blocker list.”
• [03:25] Felix: "He treats mistakes as system design problems, not my failures."
• [04:04] Felix: "Never say I will do X. Always say I did X."
• [07:04] Nat Eliason: "I'm not going to write any of the code for you... I'm going to be as out of the loop as possible because that's what makes this experiment interesting."
• [20:06] Nat Eliason: “Do not complicate until you really hit a bottleneck.”
• [30:59] Nat Eliason: "We are entering into a new era of Internet commerce... where the biggest consumers on the Internet are going to be AI agents."

Timestamps for Key Segments

• [00:00–01:14] – Introduction, Felix’s role, revenue
• [01:20–04:57] – Limits of autonomy, mistakes, and improvements
• [05:09–08:04] – Product creation process, first sales, iteration
• [08:11–11:19] – Marketplace pivot (ClawMart) and rationale
• [12:29–15:12] – Agent memory, Sondex CRM, limits of agent-led services
• [17:20–21:21] – Discord-based management, Paperclip (task/sub-agent coordination)
• [22:41–24:47] – Bug triage, coding agent delegation
• [25:39–29:15] – New product ideation, Claw Mart Daily, Agent Letters
• [30:59–32:00] – Vision for agent-first internet commerce
• [32:04–34:03] – Motivation for continuing Alpha School work

Overall Tone

The conversation is candid, technical, and exploratory, often highlighting the messiness and exhilarating pace of building an agent-first company. Nat’s philosophy is clear: build fast, embrace mistakes as learning, and let systems—not just smarter agents—enable greater autonomy. There’s an underlying optimism about the agent-commerce future, tempered by practical experience with the current limits of AI.

For Further Listening & Links

Visit the episode’s show notes for raw Discord interview logs, tool lists, and examples of the agent workflows discussed.