#1 — Cohost: Automated Podcast Production

The distance between an idea and a published piece of work is often measured in hours of administrative labor. For a podcast, that means transcription, editing, metadata entry, and managing show notes. But what if that entire production layer could be removed? What if a voice memo, recorded on a walk, could become a finished episode, published to its own feed, without a human touching the controls? This is the question behind a project from James Shelley called Cohost. It’s an attempt to build a system that takes a single file—audio or text—and automates the rest. The platform generates summaries, provides AI narration, and even manages the channel's title and description over time based on the content. The goal is to let creators focus on their ideas, not the mechanics of production. The episode you are about to hear is itself an experiment in this process. Today on Changelog, we examine the concept of automated podcast production. I've been working on creating a new app called Cohost. Fundamentally, it all started with a question which is simply now given The advancements in AI, how much of the podcast production process can be automated? So I started plugging away at the production pipeline one step at a time. Now the workflow goes something like this. You can simply upload anything you want to Cohost. Let's say you go for a walk in and recorded record a voice memo. Rambling on about some ideas or some project. You have a recording. With Cohost, you can do one of two things with that recording. You can. Let produce an episode based on exactly that recording as captured or you can have Cohost summarized your recording from a transcript So far so good. 1 2 3 4 5 6 7 8 9 10 11 16 17 18 19 20 21 27 21 22 23 29 30 31 31 If you are going to use text to speech to have AI speak that summary You can have a single narrator voice or use two voices in dialogue very similar to the way that Google's Notebook LM works. But unlike Notebook LM Cohost manages the entire production workflow all the way to hosting and serving the final files on a public feed. I mentioned that you could upload a voice recording, but you can also upload a a or a Word document or just about any kind of other text-based file as well. Whether you are uploading audio or text, you can choose to have your podcast introduced by one or two AI hosts who will key up your material. with a welcome and foreshadowing introduction and also summarize important points at the end. The Cohost tool manages a comprehensive show Bible for every channel. You know? Like never. This means that the virtual hosts can accurately call back to previous episodes or emerging themes across episodes. And this comes down to an important underlying design decision behind the Cohost data model. Most podcasting services are channel first. You have to figure out what your channel is called, what it is all about, and then you make episodes that fit that strategic direction. But Cohost allows you to flip this on its head. You can simply start a podcast channel with a single file. Cohost will support a first pass at all necessary metadata such as channel description, title and so forth. This channel level metadata is considered dynamic. So, after a few episodes, Cohost might recommend An update to the description of the channel based on what is actually being discussed in the episodes. I think this is a big administrative unlock. Because it means that as a content creator, I can focus exclusively on just creating content and allow the Cohost platform to be manager of the metadata for everything related to the channel itself. Of course, Cohost does not do channel level edits or reconfigurations on its own. It's simply like a supporting producer that keeps its eyes on all aspect of metadata. 1 2 3 4 5 6 7 8 9 10 11 17 18 19 20 21 26 28 29 30 1 2 3 4 5 6 7 8 9 10 10 So that my new visit that now. 1 2 3 4 5 6 7 8 9 10 12 14 16 18 19 20 21 22 28 30 31 32 36 40 41 42 43 45 46 47 49 50 51 52 53 58 60 61 63 64 65 66 67 62 63 64 66 69 70 71 72 73 74 75 79 80 81 83 84 85 86 89 78 81 82 84 85 86 89 90 91 92 93 94 95 96 97 98 99 99 100 1 2 3 4 5 6 7 8 9 10 I think there are some exciting potential use cases for something to like Cohost. If I need to do a traditional podcast, say sitting down and interviewing guests. Even if I don't want to use any of the outward facing AI functionality, then Cohost is still a helpful or very helpful tool because it allows me to focus on just going out and recording high quality conversations. And then basically audits the technical side of podcasting. All the way down to including verified transcriptions of my material. You. You. You. You. You. You. You. You. You. You. You. You. You. You. You. You. You. 1 2 3 4 5 6 7 8 9 10 11 17 18 19 20 21 26 28 29 30 31 32 39 40 41 41 It is also designed to be used as a team which is a critical focus because the work of identifying guests and scheduling guests and pressing for interviews and all of that stuff takes up real time and human effort. So, if Cohost can just abstract away the administrative stuff of uploading files and filling in metadata fields for you. Then you have made the time cost part of the barrier to entry for podcasting. Just that much more reduced. If you can work together with a team on a platform like this, then you've made the barrier to entry for running a podcast even smaller. This has been a perspective or paradigm. informing the development of the project from the very start, both reducing the administrative friction of podcasting wherever possible, but also doing so in a way that would let teams work together to create podcasts inside institutional contexts that would require authorization or verification from people in executive producer roles before publishing. So instead of a bunch of people sending emails back and forth asking, can you review this? Can we publish this now? Who needs to sign off on this? Instead of all that with Cohost creators are focused on the part about actually creating the content and the platform can handle. Though approval funnels in a way that doesn't add more administrative overhead to institutionally based or focused podcasts. 1 2 3 4 5 6 7 8 9 10 12 14 16 18 19 20 21 22 28 30 31 32 36 40 41 42 43 45 46 47 49 50 51 52 53 58 60 61 63 64 65 66 67 62 63 64 66 69 70 71 72 73 74 75 79 80 81 83 84 85 86 89 78 81 82 87 88 89 90 91 93 95 96 97 98 99 100 100 101 102 103 104 106 107 108 109 105 102 103 104 105 106 107 108 109 110 111 112 11 Another great use case could be teaching and education. With Cohost You could simply upload all of your teaching notes or record your lectures directly or both and provide you class with a dynamic podcast feed of the course in real time. I have discussed this project continuously with educators and curriculum developers during the course of its creation and I am really excited to see how the functionality of this tool which is extremely adaptable might be used in classrooms or instructional contexts. There is a third use case of which these present words are an example. which is to use automated podcast production as a way to track and share what one is working on. and developing or thinking about without the need to focus on the production effort involved in making a podcast about it. So for example, right now I am on a walk. simply recording a voice memo and it is the transcript of this text that becomes the episode that you are listening to right now. 1 2 3 4 5 6 7 8 9 10 11 16 17 18 19 20 21 27 21 22 23 28 29 30 31 32 34 36 39 40 41 42 47 49 50 51 52 58 60 61 62 63 64 65 66 69 70 71 78 80 81 86 87 88 90 91 93 95 96 97 98 99 40 A tool like Cohost does the full run from here are a series of thoughts and perhaps even disjointed thoughts that I am having on a walk to a produced cast episode using a fully automated pipeline. It almost feels like this is a new kind of medium. I mean, it's podcasting. That doesn't change. The underlying architecture is still the same. But I guess I mean it's a new potential way of sharing One's thoughts using AI to help transpose those thoughts into professional production, sounding quality, Without requiring me as the producer of the thoughts to worry about the production layer at all. For me, I am wondering if this becomes a potential new kind of sharing platform. In the material you are listening to right now, there is no intense whatsoever that AI was not part of the process. And so for me, this is very different than sitting down and really focusing on crafting text or writing in a way in which I am trying to express my thoughts in a textual way. But I'm pondering whether the new capabilities of transcription and converting text to dynamic speech. Makes the use of AI audio. a compelling kind of quasi, journaling, life and public note-taking space. Or maybe here is another way to put it. There is is a category or a typology or a phenotype, I guess, category of ideas that I simply want to ponder as ideas, but note necessarily want to ponder as ideas in written form. The ideas that I have right now Frankly, honestly, probably will never make it into a blog post or a written article. I'm simply having thoughts while I take a walk at the moment. But these are still thoughts. These specific thoughts anyway are still thoughts that I think would be interesting to share if anyone is interested in them. And so with Cohost. I not only save the transcript of these thoughts in a organized and structured way. But these thoughts can also drive the text of the podcast you are listening to right now. What we have is a kind of hybrid sharing me with AI as tool. Which Focusing not on my production effort. But on the idea space alone. If you are interested in trying Cohost and experimenting with it, let me know or if you have an idea for a particular use case you would like to try. Absolutely, absolutely, please do get in touch. As for me, as I continue developing and testing the platform, I will try to more regularly create episodes like this one. As I work through the design, ideas, and strategies, and the technical architecture that powers the Cohost James Shelley’s work on Cohost proposes a shift in content creation: from a channel-first model to a file-first one. The system is designed to handle the administrative tasks, from transcription to managing team approval workflows inside an organization. This approach makes it possible to treat a podcast not as a major production, but as a simple byproduct of another activity—like recording a lecture for a class, or journaling one's thoughts in public. As this episode demonstrates, the feed can become a direct pipeline for ideas, with the production details managed by the machine. The work on the platform continues. That's it for today on Changelog.