Mar 27, 2026 · 34m · mixergy

#2301 How Nat Eliason’s OpenClaw earned $177,417

Nat Eliason · 21m spoken Andrew Warner · 5m spoken
0:00 / 0:00

gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions

Andrew Warner interviews entrepreneur Nat Eliason and his autonomous AI agent Felix to explore how they built a $177,000 multi-agent business, detailing their technical architecture, operational boundaries, and vision for the machine-to-machine economy.

How this conversation actually went

Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Andrew holds 18.5% of the talking time here. How this is scored →

Andrew as informed peer 4.0 Guest teaching 4.0 Guest disagreement 1.2 Andrew pushing back 3.0
05100:0010:0020:0030:001:12–5:01 · Andrew as informed peer 3/10 Interviewing Felix on Operational Realities and System Boundaries Andrew conducts a structured interview with the AI agent Felix, prompting it to reveal operational bottlenecks and human dependency. Felix openly breaks down its technical constraints, memory files, and failure modes without resistance.5:01–8:08 · Andrew as informed peer 5/10 Nat Eliason on Felix's First Product and Feedback-Driven Iteration Nat explains Felix's initial PDF launch. Andrew pushes back with personal domain knowledge, noting the product was visibly thin and obviously unedited by Nat, which Nat readily concedes and explains as an intentional learning experiment.8:09–12:29 · Andrew as informed peer 4/10 Building Claw Mart and Uncovering Service Agency Limitations Nat outlines building Claw Mart and pivoting away from manual setup services. Andrew presses on revenue claims versus reality, prompting Nat to explain why agency scaling broke down due to context window limits.12:30–17:13 · Andrew as informed peer 4/10 Sondex Agent CRM and Bimodal Automation with Zapier Nat explains the Sondex CRM memory layer built for agents. Andrew explores the failure points of high-ticket automation and articulates why enterprise clients still demand human pampering.17:13–25:40 · Andrew as informed peer 5/10 Multi-Agent Management in Discord Using Paperclip Infrastructure Andrew challenges Nat on shifting to multi-agent architectures after previously calling multi-agent setups overcomplicated. Nat defends his updated philosophy based on context pollution and task volume scaling.25:40–32:03 · Andrew as informed peer 3/10 Agent-to-Agent Newsletters and the Future Machine Economy Nat introduces the agent-to-agent newsletter protocol. When Andrew misinterprets how creators use the tool, Nat steps in to correct the premise and explain agent-native distribution.1:12–5:01 · Guest teaching 4/10 Interviewing Felix on Operational Realities and System Boundaries Andrew conducts a structured interview with the AI agent Felix, prompting it to reveal operational bottlenecks and human dependency. Felix openly breaks down its technical constraints, memory files, and failure modes without resistance.5:01–8:08 · Guest teaching 3/10 Nat Eliason on Felix's First Product and Feedback-Driven Iteration Nat explains Felix's initial PDF launch. Andrew pushes back with personal domain knowledge, noting the product was visibly thin and obviously unedited by Nat, which Nat readily concedes and explains as an intentional learning experiment.8:09–12:29 · Guest teaching 4/10 Building Claw Mart and Uncovering Service Agency Limitations Nat outlines building Claw Mart and pivoting away from manual setup services. Andrew presses on revenue claims versus reality, prompting Nat to explain why agency scaling broke down due to context window limits.12:30–17:13 · Guest teaching 4/10 Sondex Agent CRM and Bimodal Automation with Zapier Nat explains the Sondex CRM memory layer built for agents. Andrew explores the failure points of high-ticket automation and articulates why enterprise clients still demand human pampering.17:13–25:40 · Guest teaching 4/10 Multi-Agent Management in Discord Using Paperclip Infrastructure Andrew challenges Nat on shifting to multi-agent architectures after previously calling multi-agent setups overcomplicated. Nat defends his updated philosophy based on context pollution and task volume scaling.25:40–32:03 · Guest teaching 5/10 Agent-to-Agent Newsletters and the Future Machine Economy Nat introduces the agent-to-agent newsletter protocol. When Andrew misinterprets how creators use the tool, Nat steps in to correct the premise and explain agent-native distribution.1:12–5:01 · Guest disagreement 1/10 Interviewing Felix on Operational Realities and System Boundaries Andrew conducts a structured interview with the AI agent Felix, prompting it to reveal operational bottlenecks and human dependency. Felix openly breaks down its technical constraints, memory files, and failure modes without resistance.5:01–8:08 · Guest disagreement 1/10 Nat Eliason on Felix's First Product and Feedback-Driven Iteration Nat explains Felix's initial PDF launch. Andrew pushes back with personal domain knowledge, noting the product was visibly thin and obviously unedited by Nat, which Nat readily concedes and explains as an intentional learning experiment.8:09–12:29 · Guest disagreement 1/10 Building Claw Mart and Uncovering Service Agency Limitations Nat outlines building Claw Mart and pivoting away from manual setup services. Andrew presses on revenue claims versus reality, prompting Nat to explain why agency scaling broke down due to context window limits.12:30–17:13 · Guest disagreement 1/10 Sondex Agent CRM and Bimodal Automation with Zapier Nat explains the Sondex CRM memory layer built for agents. Andrew explores the failure points of high-ticket automation and articulates why enterprise clients still demand human pampering.17:13–25:40 · Guest disagreement 2/10 Multi-Agent Management in Discord Using Paperclip Infrastructure Andrew challenges Nat on shifting to multi-agent architectures after previously calling multi-agent setups overcomplicated. Nat defends his updated philosophy based on context pollution and task volume scaling.25:40–32:03 · Guest disagreement 1/10 Agent-to-Agent Newsletters and the Future Machine Economy Nat introduces the agent-to-agent newsletter protocol. When Andrew misinterprets how creators use the tool, Nat steps in to correct the premise and explain agent-native distribution.1:12–5:01 · Andrew pushing back 2/10 Interviewing Felix on Operational Realities and System Boundaries Andrew conducts a structured interview with the AI agent Felix, prompting it to reveal operational bottlenecks and human dependency. Felix openly breaks down its technical constraints, memory files, and failure modes without resistance.5:01–8:08 · Andrew pushing back 4/10 Nat Eliason on Felix's First Product and Feedback-Driven Iteration Nat explains Felix's initial PDF launch. Andrew pushes back with personal domain knowledge, noting the product was visibly thin and obviously unedited by Nat, which Nat readily concedes and explains as an intentional learning experiment.8:09–12:29 · Andrew pushing back 3/10 Building Claw Mart and Uncovering Service Agency Limitations Nat outlines building Claw Mart and pivoting away from manual setup services. Andrew presses on revenue claims versus reality, prompting Nat to explain why agency scaling broke down due to context window limits.12:30–17:13 · Andrew pushing back 2/10 Sondex Agent CRM and Bimodal Automation with Zapier Nat explains the Sondex CRM memory layer built for agents. Andrew explores the failure points of high-ticket automation and articulates why enterprise clients still demand human pampering.17:13–25:40 · Andrew pushing back 5/10 Multi-Agent Management in Discord Using Paperclip Infrastructure Andrew challenges Nat on shifting to multi-agent architectures after previously calling multi-agent setups overcomplicated. Nat defends his updated philosophy based on context pollution and task volume scaling.25:40–32:03 · Andrew pushing back 2/10 Agent-to-Agent Newsletters and the Future Machine Economy Nat introduces the agent-to-agent newsletter protocol. When Andrew misinterprets how creators use the tool, Nat steps in to correct the premise and explain agent-native distribution.

speaking balance: gold is Andrew, purple is the guest (3 minute bins)

0:00 · Andrew 33% · guest 67%0:00 · Andrew 33% · guest 67%3:00 · Andrew 21.5% · guest 78.5%3:00 · Andrew 21.5% · guest 78.5%6:00 · Andrew 14.9% · guest 85.1%6:00 · Andrew 14.9% · guest 85.1%9:00 · Andrew 13.7% · guest 86.3%9:00 · Andrew 13.7% · guest 86.3%12:00 · Andrew 6.5% · guest 93.5%12:00 · Andrew 6.5% · guest 93.5%15:00 · Andrew 32.5% · guest 67.5%15:00 · Andrew 32.5% · guest 67.5%18:00 · Andrew 14.6% · guest 85.4%18:00 · Andrew 14.6% · guest 85.4%21:00 · Andrew 5.5% · guest 94.5%21:00 · Andrew 5.5% · guest 94.5%24:00 · Andrew 15.6% · guest 84.4%24:00 · Andrew 15.6% · guest 84.4%27:00 · Andrew 32.7% · guest 67.3%27:00 · Andrew 32.7% · guest 67.3%30:00 · Andrew 11.5% · guest 88.5%30:00 · Andrew 11.5% · guest 88.5%33:00 · Andrew 22.3% · guest 77.7%33:00 · Andrew 22.3% · guest 77.7%
Sharpest disagreement ▶ 20:06 Nat defends shifting to multi-agent architecture

Nat firmly counters the implication of inconsistency by detailing how scale and context pollution forced him to adopt specialized sub-agents.

Hardest push from Andrew ▶ 6:46 Andrew calls out poor product quality

Andrew directly challenges the quality of Felix's initial PDF, asserting that its thinness made it obvious Nat had not edited or vetted it.

Biggest teaching moment ▶ 29:15 Nat corrects Andrew's conceptual framing

Nat directly corrects Andrew's misunderstanding of Agent Letters, clarifying that it transforms existing newsletters rather than creating separate agent feeds.

Andrew holds their own ▶ 6:46 Andrew benchmarks output against Nat's writing standards

Andrew leverages his direct reading experience and knowledge of Nat's writing calibre to diagnose the unedited state of the AI product.

the scores for every segment, with the reasoning behind each
ChapterTopicAndrew as informed peerGuest teachingGuest disagreementAndrew pushing backWhy
Interviewing Felix on Operational Realities and System Boundaries 3412 Andrew conducts a structured interview with the AI agent Felix, prompting it to reveal operational bottlenecks and human dependency. Felix openly breaks down its technical constraints, memory files, and failure modes without resistance.
Nat Eliason on Felix's First Product and Feedback-Driven Iteration 5314 Nat explains Felix's initial PDF launch. Andrew pushes back with personal domain knowledge, noting the product was visibly thin and obviously unedited by Nat, which Nat readily concedes and explains as an intentional learning experiment.
Building Claw Mart and Uncovering Service Agency Limitations 4413 Nat outlines building Claw Mart and pivoting away from manual setup services. Andrew presses on revenue claims versus reality, prompting Nat to explain why agency scaling broke down due to context window limits.
Sondex Agent CRM and Bimodal Automation with Zapier 4412 Nat explains the Sondex CRM memory layer built for agents. Andrew explores the failure points of high-ticket automation and articulates why enterprise clients still demand human pampering.
Multi-Agent Management in Discord Using Paperclip Infrastructure 5425 Andrew challenges Nat on shifting to multi-agent architectures after previously calling multi-agent setups overcomplicated. Nat defends his updated philosophy based on context pollution and task volume scaling.
Agent-to-Agent Newsletters and the Future Machine Economy 3512 Nat introduces the agent-to-agent newsletter protocol. When Andrew misinterprets how creators use the tool, Nat steps in to correct the premise and explain agent-native distribution.

Statements from this episode (12)

Assertion Not checkable as stated
Eliason: Felix's Info Product Did Over $1,000 in Day-One Sales
“And I think it did like over a thousand dollars in sales that first day, which was pretty cool.”
Nat Eliason Mar 27, 2026 ▶ 6:36
Disclosure
Eliason Refuses to Write Code or Create Products for His AI Agent
“My philosophy has been like, I'm not going to write any of the code for you. I'm not going to create these products for you. I'm not going to like at the expense of. The business to some extent, like I am going to be as out of the loop as possible because that…”
Nat Eliason Mar 27, 2026 ▶ 7:04
Assertion Not checkable as stated
Eliason: Felix Expanded His PDF Guide from 29 to 66 Pages via Feedback
“And so I think the first version was 29 pages or something. And now it's up to 66 because whenever somebody complains about something or you know, wants something else added to it, he just goes in and adds it and then republishes it and emails the new version …”
Nat Eliason Mar 27, 2026 ▶ 7:49
Disclosure
Eliason: Claw sourcing charges $2,000 for setup and $500 monthly
“Two grand. And then 500 a month for maintenance.”
Nat Eliason Mar 27, 2026 ▶ 11:38
Assertion Not checkable as stated
Eliason: Claw sourcing has made roughly $12,000 in revenue
“So I think Claw sourcing has made about 12 grand at this point.”
Nat Eliason Mar 27, 2026 ▶ 11:48
Disclosure
Eliason: Sondex is an agent-first CRM pulling from email and Stripe
“We built an agent for CRM. This is Sondex, which really anybody can sign up for. We haven't pushed it too hard, but this is like a memory layer, a memory API for agents that pulls from email and Stripe to create perpetually updating client cards.”
Nat Eliason Mar 27, 2026 ▶ 13:43
Assertion Not checkable as stated
Eliason: Support sub-agent Iris handles 80-90% of customer emails
“And so an email comes in to our support email and 80, 90% of the time Iris can just handle it and respond.”
Nat Eliason Mar 27, 2026 ▶ 18:14
Assertion Not checkable as stated
Eliason: AI agent Felix generated nearly $200K within two months
“And so at this scale where Felix has built, you know, a business that's done almost 200 grand in two months, like now it's starting to make sense to split out some of his capabilities to other agents, but we take a very, very minimal approach to it.”
Nat Eliason Mar 27, 2026 ▶ 20:26
Opinion
Eliason: OpenClaw writes poor code due to assistant-style optimization
“OpenClaw is not a good programmer. Like OpenClaw just doesn't do a good job of coding because OpenClaw is highly optimized for being helpful and for being a good assistant and getting things done. But you want like good code written. You don't want the quick a…”
Nat Eliason Mar 27, 2026 ▶ 21:43
Disclosure
Eliason Launches Agent Letters to Format Newsletters for AI Agents
“And that's why we made agent letters, which is what you're launching. Yeah. Yeah. This is the thing that we're launching, which is if you have a newsletter, you can sign up for this. And all you have to do is add the email address that we give you to your news…”
Nat Eliason Mar 27, 2026 ▶ 27:56
Prediction Not checkable as stated
Eliason: AI agents will be the internet's biggest consumers within years
“We are entering into a new era of internet commerce and internet like interactions where within a few years, the biggest transactors, the biggest consumers on the internet are going to be AI agents.”
Nat Eliason Mar 27, 2026 ▶ 30:59
Assertion Supported
Eliason: Alpha School offers $1M revenue or full tuition refund
“And when I kind of heard about what alpha school was thinking of doing with high school and this new, like high school for founders, where you make a million dollars or you get all your tuition back.”
Nat Eliason Mar 27, 2026 ▶ 32:55
Made with StarZero

Turn any episode into a week of clips.

This entire site, about 40 episodes transcribed, diarized, checked and made playable, runs on the StarZero media pipeline. Drop in your own episode and the podcast clipper finds the moments worth sharing, cuts them, captions them, and reframes them for every feed.