Presented by Zapier: https://zapier.com/

Resource mentioned:
1. Tools Nat used to build Felix
2. Unedited transcript for the Felix interview
3. More
👉 All here:https://thenextnewthing.ai/nat-eliason-felix

Guest links:
👉 Nat Eliason (LinkedIn): https://www.linkedin.com/in/nateliason/
👉 Masinov: https://masinov.co

An AI agent made $177,000 running its own business—and then got interviewed about it.

In this episode of The Next New Thing, Andrew Warner does something unusual: he interviews Felix, an autonomous OpenClaw agent, before talking to its human co-founder, Nat Eliason.

Felix explains how it operates, where it’s actually autonomous (and where it’s not), and how it manages real revenue streams—from selling products to handling customer support. Then, Nat breaks down how the system works behind the scenes: how Felix launches products, builds marketplaces, manages other agents, and continuously spins up new businesses.

You’ll see how a simple experiment—“build something overnight and sell it”—turned into a multi-product ecosystem including PDFs, marketplaces, services, and agent-native tools.

The bigger idea: we’re moving toward a world where AI agents are not just tools—they’re economic actors.

⏱ Timestamps
00:00 Felix made $177K as an AI agent
00:27 Interviewing an AI agent (first ever)
01:12 Where Felix is actually not autonomous
02:24 Tools Felix runs on (OpenClaw, Claude, Discord)
03:00 Limits: memory, judgment, and calls
03:27 How Nat improves Felix through system design
04:03 Learning from real mistakes in production
05:06 First product: AI-generated PDF sold on X
06:09 $1K+ in sales overnight
07:03 Iterating products based on user feedback
08:06 Building Claw Mart (agent skill marketplace)
09:36 Why marketplaces beat service businesses
11:24 Selling OpenClaw setup services ($2K + $500/mo)
12:27 Why they paused the service business
13:21 Building an agent-first CRM (Sodex)
15:00 How agents manage customer context
17:15 Running the company entirely in Discord
18:00 Paperclip: agents managing other agents
20:15 When to split into multiple agents
22:12 Why Felix doesn’t write code
24:00 Debugging, tickets, and agent workflows
25:48 How new product ideas emerge
27:00 AI-native newsletters for agents
28:03 Agent-friendly content distribution
30:09 The future of agent-driven commerce
31:57 Why Nat isn’t going all-in (Alpha School)

👉 Join us: https://thenextnewthing.ai/

Presented by Zapier
https://zapier.com/

Episode Highlights / Timestamps
00:00 Revenue explodes after building for AI agents
00:18 The origin of Postiz as an open-source social media scheduler
01:12 Finding a “blue ocean” inside a crowded market
01:57 Adding MCP and early AI integrations
02:42 Why automation dramatically reduces churn
03:54 Growing Postiz to $17K–$20K MRR
04:03 Discovering OpenClaw and the shift toward agent-driven software
05:06 Building a CLI so agents can control Postiz
05:51 The viral “Larry” OpenClaw agent story
07:48 Why agents need strong documentation and skills
09:18 Turning a full API into a simple CLI with Claude
11:51 Why CLI tools may become the default interface for agent startups
12:45 The next startup idea: agent-native UGC video generation
13:03 Why CLI reduces token usage compared to APIs
16:21 Using Claude to build the CLI automatically
17:06 Postiz reaches $45K MRR

In this episode of The Next New Thing, Andrew Warner talks with Nevo David, the creator of Postiz, about how his revenue jumped to $45K+ MRR after a surprising shift: he stopped building primarily for humans and started building for agents.

Presented by Zapier
https://zapier.com/

Episode Highlights / Timestamps
00:00 Marketing to agents, not humans
00:45 What “agent marketing” actually means
01:30 How agents decide which products to pick
02:15 What works: clean docs, fast pages, agent-friendly content
03:54 How people are testing and tracking agent recommendations
04:48 Is SaaS dead?
04:57 Zapier’s CPTO vibe-codes a meeting recorder
05:24 Why they still won’t cancel SaaS subscriptions
06:27 When vibe coding is worth it (and when it isn’t)
06:45 Software spend vs headcount spend
07:57 The “War Council” Claude skill
08:33 How it spins up subagents + personas
09:54 How Wade built it fast using Cursor + Granola notes
11:06 Skills as a commodity vs software as a business
12:54 Using War Council for hiring decisions
14:51 Using it to analyze sales performance + feedback
16:21 Wade’s Cursor setup + switching between models
17:42 Using Codex to critique Claude when it gets stuck
18:09 How Wade structures personal context files
21:18 Building an AI chief-of-staff system
22:03 Using Zapier MCP to draft emails / run actions
24:09 Getting 800 people at Zapier using Cursor / Claude Code / Codex
25:39 Example: AI reviewing 4 massive spreadsheets fast
31:03 The “NO” hat and staying focused
32:06 Wrap

📄 War Council Skill (Claude Skill mentioned in the episode):
https://docs.google.com/document/d/1CU674IKmPCAZm2xuqMGklTA-Bq1xr1GNQW6hNydxXrE/edit?tab=t.0

Are you marketing to humans… or to agents?

In this episode of The Next New Thing, Andrew Warner sits down with Wade Foster to unpack a shift that’s already starting to change how companies grow:

AI agents are beginning to choose products on behalf of humans.

That means you may no longer be “selling to a person.” You’re trying to get ChatGPT, Claude, and other models to recommend you instead of a competitor — and the tactics are different. Wade explains what “agent marketing” actually means, what agents care about (and what they ignore), and why teams are already building tools to measure how models mention their brand.

They also tackle a question every founder is asking:

Is SaaS dead?

Wade shares an example from inside Zapier: their CPTO vibe-coded a meeting recording tool internally. It worked as a proof of concept — but they’re not canceling their SaaS subscriptions. Wade breaks down why building is cheaper than ever, but maintenance, polish, and focus are still what make commercial software worth paying for.

Then the conversation gets tactical: Wade shows how he’s using AI daily as a “second brain” inside Cursor — including a Claude skill he calls The War Council, which spins up sub-agents (ruthless CFO, wartime operator, hiring expert, design visionary, etc.) to debate decisions and return a synthesized recommendation.

This is a real look at how AI-native leadership works inside an 800-person company — without hype.

Presented by Zapier
https://zapier.com/

Episode Highlights / Timestamps
00:00 AI that runs your company
01:03 How Polsia’s agents are structured
02:33 One-click Meta ads explained
04:30 Why friction kills growth
06:18 Subscription model + nightly CEO agent
08:24 Launching multiple companies as a “fund”
10:21 Revenue split: 80/20 alignment
14:24 The Polsia economy vision
16:30 A real customer story
19:39 Should you build elsewhere first?
24:09 How Polsia grew from $20K to $600K+ run rate
25:12 The AI fundraising stunt
27:00 Live revenue dashboard explained
34:57 Live demo: launching a company
42:18 Tasks, credits, and iterations
49:30 Solo founder with AI engineers
52:12 Humans selling to humans vs agents selling to agents

In this episode of The Next New Thing, Andrew Warner interviews Ben Cera, creator of Polsia — a platform where autonomous agents build, market, and operate companies with minimal human involvement.

Polsia sets up the infrastructure (server, database, email, GitHub), builds the MVP, runs Meta ads, sends cold emails, posts on Twitter, answers support, and even iterates on product decisions.

Ben is a solo founder. Zero employees.

And Polsia is already showing a ~$600K+ run rate across subscriptions, tasks, ad usage, and revenue share — just weeks after launch.

But here’s the surprising part:

Most of the companies on the platform are only weeks old. The biggest revenue-generating startup inside Polsia is still early. This isn’t about overnight unicorns. It’s about a new operating model.

You bring the idea.
Polsia spins up the company.
You decide the budget.
The agents execute.

And Polsia takes 20% of revenue — aligning incentives with the founder.

Presented by Zapier
https://zapier.com/

Episode Highlights / Timestamps
00:00 The first billion-dollar solo company (Minecraft)
00:27 Elad’s investing track record
01:12 What “making it” really means
04:03 Where today’s “toys” become tomorrow’s giants
08:51 AI puts building power in millions of hands
09:45 Will more builders mean smaller outcomes?
13:03 AI service shops and vertical software
15:00 AI cutting permitting time from months to hours
16:39 Does AI replace CRMs and SaaS?
19:12 Is off-the-shelf software dead?
23:15 The shift from seats to AI labor units
27:36 Alexandria: translating the world’s most important books
30:36 How Elad uses AI personally
35:06 Where new AI ideas come from
37:48 What’s exciting for the next decade

“The first billion-dollar one-person company? That already happened. It was Minecraft.”

In this episode of The Next New Thing, Andrew Warner sits down with legendary investor Elad Gil — early backer of companies like Airbnb, Coinbase, Stripe, Instacart, and more — to talk about where AI is really going… and what founders are getting wrong.

Elad argues that we’re still in the early innings of AI — and that “software is AI.” The shift isn’t just better SaaS. It’s a move from seat-based software to metered digital labor. From buying tools… to buying units of work.

They discuss:

Whether “toy” AI apps can become real businesses
Why small vibe-coded projects can turn into giant companies
The agent shift (and why it changes TAM completely)
How AI eats into labor markets, not just software categories
Whether CRMs, ERPs, and landing page tools survive
Why some companies should be bought and rebuilt with AI
The real opportunity in foundation models beyond language

Elad also shares what he’s personally experimenting with — scraping and interrogating large datasets using Claude, OpenAI, and Deep Research — and why he believes the next decade will look like the early SaaS boom… but bigger.

And in a surprising turn, he talks about something very un-Silicon Valley: monuments, art, and rebuilding public beauty — including a project called Alexandria aimed at translating the world’s most important books into languages covering 80%+ of humanity.

Presented by Zapier
https://zapier.com/

Episode Highlights / Timestamps

00:00 $7M ARR as a solo founder
01:21 Profit, margins, and team size
02:51 Josh’s path from Uber to Wave
05:24 Choosing ideas in the early AI days
06:18 Why summarization felt like the killer app
08:15 Competing with Otter, Fireflies, and others
10:21 Recording real-world audio vs meeting bots
12:18 Spending more on AI to improve quality
13:39 Knowing you’re onto something from user emotion
15:09 Why Wave stayed general instead of vertical
16:12 Learning to build with ChatGPT
18:00 How Wave’s architecture evolved
19:39 Using Claude Code day-to-day
21:00 AI agents analyzing analytics and logs
25:21 The tools behind Wave (Cursor, Twilio, Adapt)
27:27 Building instead of buying SaaS tools
30:00 Using AI to ship features faster
32:06 Why Zapier matters for data portability
34:03 The future of cheap, abundant software
36:09 Running Wave like a corner store, not a startup
40:12 Growth goals without VC pressure
42:18 How Wave gets customers today
49:03 Why SEO side projects didn’t convert
50:24 “If you’re good, things might work out”
54:45 Revenue breakdown and take-home profit

What does it look like when a single founder builds a profitable AI company — alone — and quietly grows it to millions in revenue?

In this episode of The Next New Thing, Andrew Warner sits down with Josh Mohrer, creator of Wave AI, to unpack how he built a $7M ARR AI business with no full-time team — and how modern AI tools fundamentally changed what’s possible for solo founders.

Josh previously helped scale Uber in its early days, but Wave AI is a very different story. It’s a one-person, profitable SaaS built around a deceptively simple idea: record real-world conversations, transcribe them, and generate high-quality summaries people actually trust. No hype. No venture capital. No big team.

Presented by Zapier
https://zapier.com/

Episode Highlights / Timestamps

00:00 Why every email should be personalized
00:18 Ryan’s background and what Untangle does
00:45 Rethinking traditional email drips
01:12 Customizing emails based on user situations
01:39 A real example that led to a signup
02:06 Daily automated marketing insights via email
03:00 Doing things that don’t scale with AI
04:03 Walking through the AI email system
05:06 Using lead magnets and contextual data
06:09 Enriching leads and storing user context
06:45 Hourly cron jobs and email scheduling
07:39 Feeding context into the LLM correctly
08:15 Preventing hallucinated features
08:24 Sending emails with Resend
09:18 Measuring clicks instead of opens
10:12 Layering engagement-based follow-ups
10:39 Long-term personalized nurture loops
12:00 Turning marketing emails into real value
13:03 Building vertical-specific AI agents
14:15 Using Zapier and modern automations
16:12 Building systems with AI coding agents
18:27 Running multiple AI agents at once
21:27 Deciding what to build in a world of “free code”
24:09 Daily AI-generated growth recommendations
27:45 Using AI to generate and validate ideas
31:03 Increasing insight frequency, not brilliance
34:21 Why personalized email is a massive opportunity
34:48 Final takeaways

Why isn’t every email completely customized for the person receiving it — especially now that AI can do it for us?

In this episode of The Next New Thing, Andrew Warner sits down with Ryan Carson, a three-time founder currently building Untangle, to walk through a very practical, very real AI system he uses every day to grow his business.

Ryan has spent over 25 years building startups, but while setting up a “standard” email drip for Untangle, he stopped and asked a simple question: why are we still sending the same emails to completely different people? Instead of writing dozens of templates, he built an AI-powered workflow that generates fully personalized emails — based on each user’s situation, behavior, and engagement — and adapts over time.

Presented by Zapier

Episode Highlights / Timestamps

[00:00] Why Pat decided to build his own video platform after YouTube strikes
[02:06] Rebuilding a YouTube-style site in just a few hours with Claude Code
[07:30] Designing the video experience before worrying about features
[14:06] Using modern frameworks without writing code
[23:06] Adding video streaming with third-party APIs instead of building from scratch
[34:03] Letting AI debug and test the app automatically
[42:00] Deploying the app live with one command
[48:18] Why your website should be the hub, not social platforms

In this episode of The Next New Thing, Andrew Warner talks with Pat Walls, founder of Starter Story, about how he used AI coding tools to quickly rebuild a version of YouTube after his channel was hit with content strikes.

Pat walks through how he used Claude Code to design, build, debug, and deploy a working video platform in real time — without writing traditional code. Along the way, he explains why founders should treat social platforms as distribution, not infrastructure, and how owning your audience and your software changes how you think about risk, growth, and leverage.

If you’ve ever wondered how far AI can really take you in building real products, this episode shows exactly what’s possible today.