Dream it. Build it. Grow it. Sign up now and you'll be up and running on DigitalOcean in just minutes.
I’d like you to take a listen to this eleven-minute podcast; it’s a “deep dive” into JAG’s Workshop:
It might sound like a pair of professional podcasters decided to talk about this humble site, but no—it is, as you may have already sussed out, completely AI-generated. Every word, every breath, every “ummm,” “ahh,” and “like.” From voices to content to structure, it’s AI all the way down.
Last fall, the tech-adjacent world went gaga over the release of these “audio overviews” from Google NotebookLM. The idea is, you toss data (articles, URLs, audio, and so on) into NotebookLM, and Google’s AI “becomes an instant expert in those sources”, allowing you to query it for insights—or generate these two-person podcasts.
I pointed it at my site and grinned giddily as I listened to the result. Even knowing it was entirely made up, I found it remarkable for how real it sounded. I was impressed by the audio quality, the conversational style, and the “insights.” It made the site (and its author!) sound extraordinarily impressive and worth reading.
It wasn’t perfect, but I loved so much about this podcast. The opening description is an ego-stroke of massive proportions, capturing what I think is the essence of the site:
Host A: Ever get the feeling your brain just kinda bounces between totally different things? One minute you’re knee-deep in some tech spec from, like, the 90s, next you’re trying to perfect a vintage cocktail recipe?
Host B: And then bam! You’re analyzing political commentary, but through the lens of, like, The West Wing or something.
Host A: Tell me about it! Well, get ready for that times a thousand. We’re diving into JAG’s Workshop, the digital playground of this guy, Jason Anthony Guy, and trust me, it’s a wild ride.
Host B: This isn’t your grandma’s blog, that’s for sure. It’s tech nostalgia, it’s cocktail culture, it’s—honestly, it’s all over the map, but in the best way possible.
They also call me a “tech geek political junkie” and “a storyteller at heart.” It’s something I’d write about myself, were I a tad less modest. I like it so much though, I may clip some of it for my About page (it’s OK: someone else said it, not me!).
From a pure “does it sound like a podcast?” perspective, it hits all the expected notes: two surprisingly enthusiastic hosts bouncing off each other on a series of loosely related topics they awkwardly try to transition between while adding personal takes meant to sound insightful. The only thing missing was a sponsor break from Squarespace, the all-in-one platform for creating your website[1].
Using the posts on this site as their starting point, the hosts added context, connected ideas, and drew conclusions in ways that sounded remarkably human. They even added facts that were not part of the original content, but which reflected the knowledge a real person might have about the topic.
For example, when introducing my post on the death of John Amos, they say:
Host A: Remember John Amos? Played James Evan Sr. on Good Times.
I hadn’t included the character’s full name, only “Amos as James was my first TV dad”; the host added that extra detail. They also riffed on my passing references to Amos’s other roles, recognizing Good Times, Die Hard 2, Coming to America, and Psych as movies and TV shows of different genres and time periods, and used that knowledge to add a reflective coda:
Host A: It’s like how one actor’s career can have this crazy impact even across, like, completely different genres and decades.
These small moments added to the believability of “two people talking.”
I also enjoyed the range of emotion the hosts expressed: disbelief, empathy, even exasperation—which you can hear during their discussion of my The West Wing obsession:
Host A: There’s another pop culture thing this guy is all about: The West Wing.
Host B: Oh, tell me about it!
They’re also imperfect speakers, interjecting “ums,” “uhhs,” and “likes” as they speak. These disfluencies made them sound more human and conversational, and the “audio production” added to this humanness; at one point toward the end, as one of the hosts tries to remember a supposed quote (I’ll get to that later), they say “... wait, I wrote it down...,” and their voice fades off slightly as if leaning away from a microphone to reach their notes.
There are also moments that feel genuinely perceptive. At one point, they juxtapose a post about The West Wing and one about a Trump/Vance Downfall parody:
I’ve been talking about the hosts as if they’re human, because it’s hard to listen to them and think of them as AI. But AI they are, so not everything’s perfect, or even true.
There were a few speech oddities you probably caught. Multiple times the hosts spell out words, like “a-n-d” or “u-s-e-d” instead of saying “and” or “used.” They also overstate and embellish a lot in their efforts to sound erudite[2]; for example, three sentences linking to a YouTube video becomes “a whole post about the Nintendo Gameboy,” where I go “deep” about how these games “are what I grew up with” and “shaped me.” While I did admit I still have several Nintendo systems, the idea that these games shaped me is a bit of lily gilding.
And of course, being AI, they simply make stuff up. I mentioned earlier a moment when a host appeared to lean away from the microphone to retrieve a quote they attributed to me. That quote was “Life’s too short to be a snob about your passions.”
Except… I never wrote it. It’s a complete and utter fabrication. It was presented so convincingly that I briefly believed I’d forgotten writing it—but it seemed too turgid, even for me. Only a search through my entries convinced me it wasn’t mine.
They also bestow on me “degrees in both computer science and theatre arts.” While I studied both in college, I hold degrees in neither. (Indeed, I hold no degrees at all.) This invention is clearly a misinterpretation of my About page, where I write that “In college, I pursued degrees in both computer science and in theatre, television and film production.” Of course, pursued does not mean completed.
There are a dozen or more of these embellishments and outright fabrications throughout the podcast, which, taken together, help make this site (and this author) seem utterly fascinating, maybe more than is warranted[3]. But despite these examples of AI bullshit—or maybe because of them!—I found the discussion delightful.
I’m excited about the future of this technology. The quality of the podcast is already good enough to convince some that it’s two real people having an engaging discussion. My good friend Ron shared with me his experience of driving his daughter to school and listening with her to a podcast study guide for her AP Psychology class. Only after it was done did he learn from her that it was AI generated.
A former co-worker uses this technology to create podcasts on a wide range of topics. He spends a few hours on a weekend filling out spreadsheets with sources of information, pours them into NotebookLM, adds intros and music, and uploads the results to Apple Podcasts and Spotify. The results are credible and riveting. (No link because he doesn’t disclose it’s AI generated, and I don’t want this to be the thing that outs it!)
Over the next couple of years, the audio will only improve; it’s already nearly indistinguishable from human voices today, though NotebookLM is currently limited to just two English-speaking, American-accented voices. Other companies offer more voice options—accents, languages, tonality, and so on. I expect those to eventually find their way into NotebookLM.
I also anticipate more control over the kind of podcast (or more broadly, audio) that’s generated. A one-person podcast, perhaps; or a panel discussion where one person acts as host to a team of opinion-makers; maybe even a round-table style discussion where each person brings a topic and gives their takes. Video would be a natural next step.
And, of course, some much-needed improvements to the AI’s “intelligence.” Inaccurate summaries and hallucinations continue to be among the biggest issues in the industry—especially when presented as fact—and it’s not clear how long it will be before these issues go away.
Despite this—and the many other issues AI has—I’m excited about the game-changing potential of using AI as a personal knowledge base and information retrieval system, especially when paired with audio and video as the method of sharing the results.
This space continues to fascinate me.
Marques Brownlee (MKBHD) tackles the eternal product marketing question of “Product or Feature” when it comes to AI.
He highlights the Humane Ai Pin and Rabbit R1 as examples of “AI as a product,” that is, a standalone thing you buy that “does AI,” versus being integrated into existing products, such as a phone or smart home device.
Both the Humane Pin and Rabbit R1 require AI for them to be useful. Without AI, they are merely interesting hardware products that provide no meaningful functionality. Everything they do with AI can easily be (and perhaps already is being) done on a phone, either via an app or integrated into the system. The unique hardware bits they have exist solely because they’re not phones and need some mechanism for us to interact with them. A phone without AI is still useful. A Humane Pin without AI is… what? An attractive accessory?
AI is an enabling technology. People aren’t buying AI. They’re buying what AI can do for them. The question for any company selling an “AI product” is “to do what?”
If the what can be incorporated into another, more popular or widely used product, that “AI product” is in danger of being obsoleted, or at least becoming less relevant.
Marques uses the once-popular Clubhouse to highlight this transition. The enabling technology for Clubhouse was “live voice chatrooms”. However, the product was effectively obsoleted after other companies incorporated this feature into their own, more popular apps. People didn’t want Clubhouse, per se. They just wanted to chat with each other.
It’s likely we’ll see this happen a lot more for “AI products.” As Marques points out, Apple’s WWDC Keynote showcased many features powered by AI. In almost every instance, products that do “that thing” already exist today. Such products may soon become superfluous to a large number of Apple’s customers, because customers care about the feature, not the product.
One example: The next versions of iOS and macOS bring the ability to “rewrite, proofread, and summarize text,” directly integrated into just about every text field on the systems. This is huge for Apple customers. But if you’re, say, Grammarly, for whom “AI writing partner” is what you sell, you’re now competing with Apple in your only business. Feature, not product.
This isn’t to say Grammarly goes away. They provide some features not offered in Apple’s version, and as they noted to NPR:
Whenever new entrants come into our market, the reality is that we see increased demand for Grammarly.
OK, sure. But how many people will pay $12/month when they can get a basic “AI writing partner” for free?
As Apple demonstrated, it’s possible to build your own AI stack, and subsume AI-powered features, making them “just another bullet point” on a deck or press release.
Look at ChatGPT, for instance. Despite being the best-known, most popular AI chatbot, it was the last item mentioned in Apple’s two-thousand-word Apple Intelligence press release. And it won’t even be the only chatbot available on Apple’s devices. Eventually you’ll be able to replace ChatGPT with Anthropic’s Claude, Google Gemini, Microsoft Copilot, and who knows what else. Feature, not product.
And dollars to doughnuts, Apple already plans to Sherlock all of these chatbots, as they do with so many other features-pretending-to-be-products.
So is AI a Product or a Feature? Marques demurs:
Now there is no answer yet as far as is AI actually a feature or a product, which one will win, which one will lose, we don't know. But I think if history is any indication, I do think that more people in the long run are going to end up using this AI stuff as a feature, more than going to, like, a standalone thing.
I agree that for most people AI will simply be the technology that powers features they love. They won’t go seeking out AI, and won’t buy standalone “AI products”. In fact, most won’t think about AI at all.
The trajectory technology takes is to transition from a visionary product to an important feature of another product to a mundane feature used without thinking.
A great example comes from my friend Ron Lue-Sang: Biometric authentication.
In popular media, fingerprints, facial recognition, and retina scans once demonstrated just how advanced the civilization (or organization) was. Visionary.
Then biometrics were added to products like door locks as a key reason to buy them. Important.
Today we unthinkingly unlock our phones and computers with our fingers and faces. Mundane.
Companies will continue to sell “AI products” for as long as they can, and many of them will exist as foundational technologies and find ways to differentiate themselves, but most are on a glide path toward irrelevance. For me, and I think for most people, AI is absolutely a feature, not a product, and one we won’t even think of as AI.
So perhaps the question isn’t “is AI a product or a feature?”
Perhaps it’s “when will AI become mundane?”