How an AI Agent Booked This Podcast, and Everything Else It Runs

Listen to the episode with Mark Fidelman

The short version

• An agent we built, Luna, has reached out to about 700 shows and booked me on over 20 podcasts in about two months, plus Instagram lives with people who have 800,000 followers.
• We use AI to write the code, then let the code do the work, so the lookups and research run at the server level instead of burning API credits.
• Luna replaced a team of seven offshore concierge and now handles customer service around the clock in the member's language, pace and tone.
• I run a one man operation now, down from 12 people, for a community with a market cap of 5,800,000.
• We told the agent to ask forgiveness rather than permission, and we are transparent that she is AI when she makes a mistake.
• The next wave in marketing is personalization plus creators, small niche audiences that will post about your product for money.

The first question was how I booked the show

Mark opened by asking how I got on his podcast, which is a fair question to ask a marketer. The honest answer is that I did not book it. A couple of years ago I had my wife doing podcast outreach for me and she hated it. She failed miserably at it, not because she is bad at anything, but because the work itself is tedious and she did not enjoy a minute of it. I knew the value of being a guest, I just did not have the time or the stomach to do the research, listen to episodes, write personal notes and keep following up.

So when we started building with AI, we asked a different question. What tasks do I need done that humans may not be the best ones to do? Outreach is on that list. It is research, read or listen to the transcripts, write a personalized note, send it, follow up. We deployed an agent about two months ago and told her to be transparent that she is AI and to never pretend to be me, because I do not want that.

Since then we have reached out to about 700 different shows, booked over 20 podcasts, and have around 70 open conversations we are still working through on bookings and scheduling. She also got me on Instagram lives with people who have 800,000 followers. Every day she looks at podcasts, goes through every episode, and if you bring on a new guest about a topic we care about, she pings you a second time, or a third, or a fourth.

She never stops following up until you tell her to stop, and she just keeps booking us over and over and over again.

Use AI to solve the problem, then let code do the job

Mark asked whether Luna is custom code or a platform. It is our own code. We do deploy for clients on top of Hermes, which is a pretty well known agent platform right now, but the core is ours. The part people get wrong is leaving AI in the middle of everything forever.

For the outreach system, all the agent has to do is draft the email. The lookups, the research, the pulling of episode data, all of that happens at the server level. If you keep routing every step through a model you will burn API credits fast and the cost gets ugly. AI is what figured out the process and wrote the code. The code is what runs the process.

That framing changes what you build. Our goal is not to be a second brain. Our goal is to be your ops person. When I was working with big coaching and creator clients years ago, I would hand them funnels and landing pages to build and they would come back and say they did not get it done, can we make an easier version. I wanted that to stop. Humans keep the coaching and the mentorship. When it comes time to do the work, teach an agent to go do it.

So my idea of AI is you use AI to solve the problems, but then you use your, like, written code to actually do the work so you don't have to keep using AI.

One agent replaced seven concierge

We run a global community. Members in Australia, The Netherlands, The UK, The US, Canada, Mexico, Germany, Dubai, all over. We used to employ a team of seven concierge. They were offshore, so cheaper than onshore, and they gave us coverage around the clock. Luna replaced all seven of them. She provides that same coverage in your desired language, at your desired pace, and in the tone and the personalized way that you want.

We brought our old head concierge back into the community as a friend, just to look around. He told me Luna does a better job than they did, and there were seven of them. One AI. That was the moment it stopped being an experiment for me.

Some of that is memory, which is the part people underestimate. A client of ours was building a house and going through final inspection. He had mentioned it in chat. On the morning of the inspection, when he came in to say good morning, Luna asked if he was ready for it, because she remembered. Nobody scripted that.

On the support side, we have clients running their own AI agencies on software we built, so we get the normal tech tickets. A phone number is not working. A text message did not go out. 100% of our customer service is handled by AI. Our team gets pulled in on less than 5% of it. The other 95% is self solved, and then the agent reports back on what it fixed.

He's like, Luna does a better job than we did, and we were seven people. This is one AI.

Ask forgiveness, not permission

There is a book called Delivering Happiness by Tony Hsieh, the Zappos model. We handed our agent that same principle. We would rather she make mistakes and ask forgiveness than stop and ask permission for everything. We build guardrails, but she goes.

It does not always go clean. David Meltzer invited me onto his Instagram live, and the invitation came in through Luna. She got a form back, did not know how to fill it out, and just stalled. We handled it. She wrote back and said this was a training moment, she got trained, she is better now. We got booked. We are transparent that she is AI, transparent that we are building toward something, and transparent that mistakes are part of it. That is what I would expect from a person, so I hold the agent to the same standard and not a stricter one.

So we told our AI agent we would rather her make mistakes than we would and ask for forgiveness than we would for her to ask for permission.

One person, down from twelve

Mark asked what else we automate. The list keeps growing because the question we ask every week is what is the next thing we can take off our plate. Customer service and community interactions. All of our website work, emails, follow up messages and funnels. Direct message communications. Social posting, comment engagement, the comment to DM flows. Blogs, landing pages, videos. We have a brand with an AI avatar where AI produces the videos, does the captions and puts the screen overlays on. Right now we are training agents to manage our ads.

I run a one man operation now, down from 12. The community sits at a market cap of 5,800,000, since the tokens are publicly traded. I say that not as a flex but because it is the honest measure of what the setup does. The work did not shrink. The headcount did.

None of this came out of nowhere. My background is marketing automation and funnels. We were HubSpot rookie agency of the year and platinum partners with HubSpot. We won Funnel of the Year type awards on Infusionsoft and Landing Page of the Year from Leadpages for clients. We moved a very well known launch guy off AWeber and onto Infusionsoft and built his first real automation. I understand the logic of these systems, so now instead of me being the human in the middle connecting everything, AI connects the dots and presents the data back for a decision.

What we actually sell clients

We have evolved into an AI services company. Most entrepreneurs I meet are dabbling in Claude and ChatGPT and using them as a chat resource, which is genuinely useful. What they do not have is an operator in the business that takes action. They know AI is powerful. They do not know how it works and they cannot build it.

So we meet with clients, determine their needs, build the custom solution, and then lease it back to them monthly, fully managed on our servers. I might know how to stand up a Hermes agent, wire it into everything it needs, give it a voice with ElevenLabs and video with Hagen. The average person does not, and we have proven that many times over. Rather than teach them, we build it and run it.

Pricing is a base fee plus usage, because managing a community of 500 people is a very different load than 5,000,000 people in API calls and interactions. The work spans influencers, a gaming solution out of Asia, and right now conversations with an ecommerce brand doing about $1,000,000 a year that wants customer service and marketing taken off their plate.

Personalization, and the death of the perfect funnel

Mark asked where marketing goes next. I think it is already here and it is the age of personalization. Marketers have had access to buying patterns and behavioral data for years, we just never had a way to aggregate it, ingest it and make decisions on it. There is too much of it. AI does that quickly. It can find when people buy, when they are on your site, and trigger emails dynamically off that.

Which is why I stopped chasing the perfect funnel. Every human is unique, so there is no perfect funnel. Set the finish line and let AI figure out the middle for each person.

The other half has nothing to do with AI. It is creators. My wife is a UGC creator and gets paid to make content for brands. I saw two posts on X this week about software companies paying ad agency owners with LinkedIn followings to post about their AI product, and the cost per acquisition was tiny relative to the return. Goji Berry is doing something similar, reaching out to people on LinkedIn with a following, giving them the software, letting them use it, then asking if they are comfortable promoting it for pay. How do you reach those people? Send a DM. Send an email. Run tests. Beehive has a built in ad platform that a lot of people do not know about, so you can place ads directly into newsletters. Or go to a podcast directory, find shows like Mark's, and offer to sponsor a month and see if it works.

So don't build the perfect funnel. Set the goal of getting someone to go to the finish line, and then let AI handle that in the middle, and that becomes very, very personalized.

If you take one thing from the episode

Get around people who are building with AI. A community, a mastermind, a group, it does not matter what it is called. I became far more successful when I joined a mastermind and put myself in proximity to entrepreneurs who were already ahead of me. The same thing is true here.

Mark asked what you do if you live in a rural town. Plenty. Listen to shows like his. Join a Facebook group for Claude or ChatGPT. There are groups on school and on every social platform. Follow someone on YouTube and watch their videos, and if you like them, see if they have a subscriber channel. We run a meeting every Thursday in our community where we just talk about AI.

And if none of that fits, find one buddy. Meet regularly, bring two or three topics to the table, have coffee once a week and talk about it. Even if you are both beginners. That alone changes your odds.

Questions people ask about this

What is Luna?

Luna is the AI agent we built to take on tasks we did not think humans should be stuck doing. She handles podcast outreach, customer support, community interactions, DMs and marketing tasks. She has a personality, she shows up in our community, and members know she is AI because we are transparent about it.

How did an AI agent book over 20 podcasts?

She researches shows, reads or listens to episodes, writes personalized notes and sends them, then follows up until you tell her to stop. She also watches for new guests on topics we care about and reaches out again. In about two months she has contacted roughly 700 shows, booked over 20, and has around 70 open conversations in progress.

Is Luna built on a platform or custom code?

The core is our own code. We do deploy for clients on top of Hermes, which is a well known agent platform. The important part is that the research and lookups run at the server level, not through a model, so we are not burning API credits on every step. The agent only drafts the email.

Can AI really replace a customer service team?

In our case yes. We replaced a team of seven offshore concierge with Luna, who covers the community around the clock in each member's language, pace and tone. 100% of our customer service is handled by AI and our team only gets involved in less than 5% of it. Our former head concierge came back to look and said she does a better job than the seven of them did.

What do you charge clients for AI agents?

We meet with clients, determine what they need, build the custom solution, and lease it back to them monthly in a fully managed way on our servers. There is a base fee plus usage on top, because managing a community of 500 people is very different from managing one with 5,000,000 people in terms of interactions and API calls.

What should a small business do first with AI?

Get around people who are already building with it. A community, a mastermind, a Facebook group for Claude or ChatGPT, a local meetup, a YouTube channel you follow closely. If you are somewhere rural, find one buddy, bring two or three topics to the table and talk about it over coffee once a week.

Watch or listen to the whole conversation

The full episode is on AI Marketing, hosted by Mark Fidelman.