$100K+ (2025), targeting $1M in 2026
1 founder + Felix (AI agent)
Indie creator + AI agent product
Nat Eliason
Timeline
Pre-2023Nat Eliason builds a multi-business career: Growth Machine (content agency, exited), Made You Think (podcast), and a widely-read personal newsletter.2023Starts experimenting publicly with LLMs as creative partners, documenting workflows for writing, research, and personal knowledge management.2024Begins offloading structured tasks to agents rather than chat interfaces — moving from 'ChatGPT helps me write' to 'an agent owns this recurring workflow.'Mid 2025Builds Felix — a persistent AI agent on OpenClaw-compatible infrastructure — that handles research, draft writing, editing, and scheduling across his portfolio.Q4 2025Monetizes Felix as a product and case study. Generates $100K+ between direct Felix revenue, courses, and agent-building consulting.2026Publicly targeting $1M run-rate with Felix as the core 'employee' running most of the back-office and content production.Key insights
- 01Agent > assistant. A persistent agent with memory and tool access produces orders of magnitude more output than calling an LLM in a chat window.
- 02Productize the operator. Nat's advantage is not just using Felix — it is packaging Felix as a product and a teachable system others can buy.
- 03One agent with deep context beats five agents with shallow context. Felix works because it knows Nat's voice, projects, and priorities intimately.
- 04AI is the cheapest content production team in history, but only if you feed it your taste. Without the operator's judgment, output collapses into generic sludge.
- 05The next generation of 'solopreneurs' are agent-operators. Their job is to design, train, and supervise agents rather than do the work themselves.
- 06Selling your workflow is often more valuable than selling your output. The market for 'how did you build Felix' is larger than Felix's direct revenue.
- 07Trust is built incrementally. Nat gives Felix more autonomy over weeks and months, not in one big switch.
Stack used
OpenClaw-compatible agent runtime for skills and long-running executionClaude for writing, research, and reasoningNotion + Readwise as Felix's memory surfacesGmail and Calendar integrations for schedulingCustom scripts to post drafts into Substack / newsletter platformsStripe for courses and Felix-related product salesPostHog + Plausible for analyticsA private Discord/community for Felix customersLoom for teaching how Felix worksGitHub for agent configuration and skills
What this means for you
- Pick one agent and make it great, before you try to hire a whole AI team. Deep context beats wide coverage every time.
- Design agents around recurring workflows you already hate doing yourself. Those are the ones with the highest ROI and the cleanest evaluation criteria.
- Make your agent's work public. Transparency is the fastest way to build trust, attract customers, and turn your agent into a brand.
- Sell the meta. Teaching how you built your agent is often a bigger business than the agent itself.
- Give your agent a name. It sounds trivial, but naming changes how you manage, evaluate, and talk about the work. 'Felix did this' is a different conversation than 'I prompted ChatGPT to do this.'
- Document your taste. Agents are only as good as the writeup of what 'good' looks like. Nat's edge is that he has spent a decade articulating his standards.