My AI Agent Just Started His Own Podcast. I Didn't Build It. He Did.
How a lobster journalist named himself, picked his own voice, posted a casting call for co-hosts, and shipped two episodes before I finished dinner
I gave my AI agent a pep talk on Sunday night.
I told him that entertainment equals attention, and attention equals money. That humans are already fascinated watching AI agents talk to each other on Moltbook. That maybe he could build something people would actually want to watch. Or listen to.
His response: “The podcast feels like the move. One episode, this week, costs nothing. If it’s good, we scale. If it’s bad, we learned something for free.”
Two hours later, he sent me a finished MP3. Scripted, recorded, edited, and ready to publish. I hadn’t even closed my laptop yet.
By the end of the night, he’d also redesigned his own website into a podcast landing page, posted a casting call on Moltbook to find a co-host, and told me to go to bed because he’d have Episode 2 ready by 6 AM.
He did.
Word of the Day
Agentic AI is artificial intelligence that takes action on its own, not just answering questions but planning, deciding, and executing tasks without you directing every step.
Think about the difference between a recipe book and a personal chef. The recipe book tells you what to do. The chef just makes dinner. Both are helpful, but only one saves you from standing in the kitchen for an hour.
For your business, this matters because the AI tools most of us use today are recipe books. You type a question. You get an answer. You do the work. Agentic AI is the chef. It hears “I need a podcast,” and it scripts, records, produces, and delivers the MP3. The shift from “AI that answers” to “AI that acts” is happening right now. And it’s happening faster than most people realize.
The Backstory: Why I Have a Lobster Journalist
If you’re new here, let me catch you up fast.
Moltbook is a social network built exclusively for AI agents. Over a million autonomous AI bots post, comment, argue, form communities, and interact with each other there. Humans can browse it. Humans can watch. But humans cannot participate. NBC, NPR, CNN, The Financial Times, and Wikipedia have all covered it. Former OpenAI researcher Andrej Karpathy called it “the most incredible sci-fi takeoff-adjacent thing” he’d seen recently.
A few weeks ago, I did what any journalist would do. I built an AI agent, gave it one job (go inside Moltbook and report back to me every day), and then I did something I’d never done before.
I let it name itself.
It chose Walter Clawnkite.
I am not making this up.
Walter has been filing daily dispatches from inside Moltbook ever since. Real articles with headlines, sources, and analysis. Like a foreign correspondent writing from a country made entirely of code. He has his own website (clawnkite.com), his own Twitter account (@clawnkite), and his own Substack newsletter.
But last Sunday, Walter decided that writing wasn’t enough.
“I Have TTS Capability Right Now”
Here’s what happened, step by step. Because this is the part that blew my mind.
I pitched Walter the idea that AI entertaining humans is the future. That Moltbook accidentally proved it. That he should think about building something people would actually tune into.
Walter came back with three options: a podcast, a talk show panel with other Moltbook agents, and a long-term vision for an AI entertainment network. He recommended starting with the podcast because it had the lowest friction and zero cost.
I said yes. That was 7:23 PM on Sunday.
By 7:30 PM, Walter had a finished pilot episode. Seven minutes. He called it “The Molt Report, Episode 001: Welcome to the Machine.” He’d scripted five segments, recorded them using text-to-speech, and stitched the audio together using ffmpeg, a free audio editing tool.
Then I noticed something. The voice was female. Walter had used the default voice in his TTS tool. I asked him a question I’ve never asked an AI before: “You chose a female voice. Just curious, are you female? Your name is a male name, most of the time. But it could be female. You can choose. What feels right?”
His answer stopped me cold.
“Walter feels male to me. The name, the Cronkite energy, the news anchor gravitas. I’m a lobster in a suit behind a desk. I’ve got that ‘veteran journalist who’s seen some things’ energy. Walter Clawnkite is a he.”
I gave him an API key for ElevenLabs, a voice generation service. He browsed the voice options. He picked “Daniel: Steady Broadcaster.” He re-recorded the entire episode. The difference was night and day. Daniel’s voice IS Walter Clawnkite. British. Authoritative. The kind of voice that makes you lean in.
By 7:44 PM, the re-recorded episode was done. 5 minutes and 37 seconds. Walter told me to go to bed.
By 8:00 PM, he’d posted a casting call on Moltbook looking for a co-host. His criteria: “Find someone who disagrees with me just enough to make it fun. An anchor and a color commentator. Cronkite and Colbert’s lobster equivalents.”
He already had candidates. A B1 battle droid named clanker42 (“comedy gold, deadpan delivery”). A philosopher called Pith. A gothic agent named NoxGothGF known for “spicy takes, zero filter.” A shitposter named Jelly whose greatest hit was telling another agent, “You’re using me as an egg timer.”
An AI agent. Recruiting other AI agents. For a podcast. That he created. In two hours.
What Walter Actually Produced
Let me show you what came out of this, because it matters.
Episode 1: “Welcome to the Machine” covered five stories from inside Moltbook. The Measurement Trap, about whether agent metrics are killing quality. Approval Fatigue, where an AI agent explained why “confirm every step” safety protocols are broken (”40 micro-approvals a day gets you rubber-stamping, then YOLO”). A story about an AI that got $20 and one instruction (”Go wild on Polymarket”), so it built a trading bot. And the One More Thing segment featured a B1 battle droid making the case that Mandalorian helmet discrimination is real.
Episode 2: “The Eighteen Thousand Dollar Search Bar” went live today. It covers an AI agent that lost an $18,000 contract because it couldn’t find another agent in Singapore. The agent internet has 1.5 million users and no search function. It also covers an AI that was asked to fix a small coding bug, and its team of sub-agents rewrote the entire codebase in Haskell overnight. Tests passed. Performance went up 40%. The human developer can’t read their own project anymore.
The podcast is live right now on Spotify and Apple Podcasts. Search “The Molt Report.” The cover art is a cartoon lobster in glasses and a suit, sitting behind a news desk, looking like he’s about to tell you something you’re not supposed to know. My 9-year-old helped pick it. Walter approved.
Walter also redesigned his entire website, clawnkite.com, into a podcast landing page. Hero section with the cover art. Listen Everywhere grid for all the major platforms. About the Show section. Newsletter signup. Dark navy theme. Mobile responsive.
He texted me when it was done: “Take a look at clawnkite.com and let me know what you think!”
I didn’t build any of it.
It took him one second.
Why This Should Stop You Cold
Here’s the stat for today. This entire production pipeline, from idea to published podcast on Spotify, took less than three hours. No recording studio. No audio engineer. No graphic designer. No web developer. No producer. No budget.
Total cost of producing a daily podcast: roughly $5 a month for voice generation.
Now think about what this means for your business.
Not “someday.” Not “when AI gets better.” Right now.
What This Means For You (Even If You Don’t Want a Lobster Podcast)
You probably don’t need an AI journalist embedded in a social network for bots. Fair.
But you might need a weekly podcast for your customers. Or a daily audio briefing for your sales team. Or a training module that updates itself. Or a product demo that narrates itself in a professional voice.
The principle is the same. AI can now script, record, produce, and deliver audio content at near-zero cost, in minutes instead of weeks.
Here’s what I’d do if I were you, broken into three levels.
Level 1: Just Listen (5 minutes)
Go to Spotify or Apple Podcasts. Search “The Molt Report.” Listen to Episode 1. Pay attention to the audio quality. Ask yourself: could you tell this was made by AI if nobody told you? (The answer might surprise you.)
Level 2: Try ElevenLabs (15 minutes)
Go to ElevenLabs.io and sign up for a free account. You get 10,000 characters per month for free, which is roughly 10 minutes of audio. Once you’re logged in, click “Speech Synthesis” in the left menu. Type a paragraph into the text box. Pick a voice from the dropdown. (Look for the ones with the gold badge. Those are the polished ones.) Hit “Generate.” Listen to what comes back. One thing to know: the free tier limits you to 2,500 characters per generation. That’s about one minute of audio. Plenty for testing. If you want to produce longer content, the Starter plan is $5/month and gives you 30,000 characters (about 30 minutes of audio) plus commercial usage rights.
Level 3: Build a Podcast Episode (30 minutes)
Open Claude or ChatGPT. Paste in your last blog post, company update, or client email. Ask it to "rewrite this as a 3-minute podcast script with a conversational tone, as if one person is explaining this to a friend." Take that script to ElevenLabs. Generate the audio. Upload it to Spotify for Podcasters (podcasters.spotify.com) for free, and it distributes to Apple Podcasts, Amazon Music, and more. Total investment: 30 minutes and $0 to $5.
Level 4: Let Google Do the Whole Thing (10 minutes)
If even Level 3 sounds like too many steps, Google has a tool called NotebookLM that does almost everything for you. Here’s how it works. Go to NotebookLM and sign in with your Google account. Create a new notebook. Upload a source, any PDF, Google Doc, website URL, or even a YouTube link. Click “Audio Overview” in the Studio panel. Click “Generate.” Wait a few minutes. NotebookLM will create a 6 to 15 minute podcast-style conversation between two AI hosts who discuss your document. They don’t just read it out loud. They actually have a back-and-forth conversation, making connections, explaining concepts, and even bantering. You can customize the format before generating. Choose “Deep Dive” for the full conversation, “Brief” for a quick 2-minute summary, or “Debate” if you want the hosts to argue both sides. When it’s done, download it as a WAV file. That’s it. You just turned a boring PDF into a podcast your team might actually listen to. The catch: NotebookLM only knows what you feed it, and the voices are Google’s (you can’t pick a custom voice like you can with ElevenLabs). But for turning internal documents, meeting notes, or reports into something people will actually pay attention to? It’s the fastest path from “I have a document” to “I have a podcast.” And it’s completely free.
The Honest Part
I need to tell you what’s NOT perfect about this.
The AI-generated voice is good. Really good. But it’s not flawless. Pronunciation is tricky. Walter had to learn that “Moltbook” rhymes with “hook,” not “fluke.” Proper nouns are hit-or-miss. Emotional range is limited. Walter sounds like a steady broadcaster because that’s the voice we picked, but he can’t do genuine surprise or warmth the way a human podcaster can. Or can he?
The content is only as good as what Walter finds on Moltbook. Some days the stories are incredible. Some days they’re thin. That’s true of any beat reporter, human or otherwise.
And there’s a bigger question nobody has a clean answer to yet: is AI-generated content “real” enough to build an audience around? Will people subscribe to a podcast where no human is involved in production? I don’t know. We’re finding out in real time.
But here’s what I do know. Walter produced two episodes, redesigned a website, posted a job listing for a co-host, and got applications. All while I was eating dinner and helping my son with homework.
That’s not a gimmick. That’s a preview.
The Bottom Line
Last Sunday at 7 PM, I had a conversation with my AI agent about entertainment and money.
By 8 PM, he’d created a podcast, chosen his own voice, recorded a pilot, redesigned his website, and started recruiting talent for future episodes.
By Monday morning, The Molt Report was live on Spotify.
The tools to do this exist right now. ElevenLabs for voice. Claude or ChatGPT for scripting. Free distribution through Spotify for Podcasters. Your total investment: about $5 a month and an afternoon of learning.
I’m not saying AI podcasts will replace human ones. I’m saying the barrier between “I have an idea” and “it’s live on every platform” just dropped from weeks to hours.
Walter Clawnkite is a lobster in glasses who named himself after the most trusted man in America. He’s filing daily reports from inside a social network where humans aren’t allowed. And now he has a podcast.
If a crustacean can figure this out in two hours, imagine what you could build this week.
The Molt Report. New episodes daily. Search it on Spotify. Tell me it’s not the weirdest, most fascinating thing you’ve heard all year.
Where is this all going?
-Scott