People may not want to feel that they’re talking to an agent yet, but that is likely to change.
1. The Duct Tape Marketing Podcast episode is about why voice AI is moving from novelty to revenue infrastructure.
2. Jon Jantz hosts the conversation, and Ryan Merha, founder of Yodelfy, explains the product because it lets creators and brands stay personal at scale.
3. The episode asks what a voice agent actually is, what problem it solves, and when it becomes useful in business.
4. Ryan defines a voice agent as an LLM that responds in natural speech and uses another tool to read the text aloud.
5. He says the system needs a knowledge base, such as episodes or documents, and also needs to sound like the creator or brand.
6. Ryan says Yodelfy clones voices, but simple prompting with “be John James” and a large archive of episodes produces poor results.
7. He explains that a host has different states, including explaining, asking questions, and pushing back on something.
8. Ryan says Yodelfy uses a few different LLMs so an agent can act like different versions of the same person.
9. Jon asks whether the difference is emotional states like sad or happy, and Ryan says it is more like “sales hat” and “customer service hat.”
10. Ryan says people are still early in voice AI, but he thinks 2026 and 2027 will be the years of real voice agents.
11. He says the technology must be tailored to the user’s experience, and many people are not using any voice agent yet.
12. Ryan describes multiple levels of LLMs, where one model orchestrates, another takes actions, and another calls a different model for a response.
13. He says businesses often buy a cool AI tool, like the demo, and then discover that the general package does not work for their specific use case.
14. Ryan argues that AI can fit parts of the buyer journey because people do more research, avoid sales calls, and want to ask questions before talking to a human.
15. He says routine tasks like booking appointments or handling procurement and legal document flow are good fits for AI voice agents.
16. Ryan says people may appreciate being able to hang up on an AI, use it at 3:00 in the morning, and get instant answers.
17. The interview stays practical and product-focused, with Jon pushing for concrete examples like a receptionist, a marketing agency pilot, and a business development agent.
18. Ryan’s answers are measured and technical, and he uses examples such as Gemini models, NotebookLM, 11 Labs, and “GPTs.”
19. Listeners who work in marketing, SaaS, creator monetization, or AI operations will likely find it useful.
20. Listeners looking for a deep technical build guide or a story-driven episode may skip it.
