Sierra: Huxe was this AI audio app, right? You'd open it up in the morning, hit play, and it would give you a personal podcast, a briefing on your day, your meetings, important emails, even the news.
Marcus: Wait, so it's like a personalized, uh... what do you call it, like— an NPR for your life, but automated?
Sierra: Totally, right? And it was built by ex-Googlers, these super smart folks. They raised, like, four or five million bucks. They launched. And then, eight months later, they just—
Marcus: Eight months?
Sierra: Yeah.
Marcus: For how much?
Sierra: Like four and a half.
Marcus: No way. What happened, the app was bad?
Sierra: That's the thing! It wasn't that the app was bad. The article says they just got rolled up.
Marcus: Rolled up? As in...?
Sierra: Yeah, like—beat—the same month Huxe announced it was winding down, Spotify and Amazon announced they were pushing deeper into that exact same personalized, AI-generated audio space.
Marcus: Oh, so they just couldn't compete, right? Too small.
Sierra: Well, it's more nuanced than that. The big insight here is that for consumer AI startups, especially, the biggest risk isn't even finding a market.
Marcus: Wait, not finding a market? That's what I thought it would be.
Sierra: Nope. It's actually building something so good that a platform giant just decides to roll it out as a minor feature in their own, already massive, product.
Marcus: Man, that's gotta be soul-crushing. To build something, raise money, get it out there, and then watch Spotify just drop an update that does basically your whole business model.
Sierra: Right? It's... it's just messed up. Because Huxe, they needed you to like—you had to download their app, right? A whole new app.
Marcus: Yeah.
Sierra: But Spotify, they're already there. Google's already got your email, your calendar. They just... they just push an update, say 'Hey, personal podcasts!' or 'NotebookLM Audio Overviews' and it's just built in. No new download, no nothing.
Marcus: It's like that amazing third-party calendar app I used for years. It had this one killer feature for scheduling, like, it was perfect. Then one day, Apple just built that exact function into the native iOS calendar app, and that startup was gone six months later. It feels inevitable sometimes.
Sierra: Yeah, that's exactly the kind of thing. And it's not just the distribution.
Marcus: Right. It's also, what, like... finding users?
Sierra: No, it's more about—it's a trust barrier. The article points it out. Huxe needed access to your email and calendar to work best. And that's a huge ask for a new app, right? But you've already given that trust to Google or Microsoft.
Marcus: True. I'm not giving some random app my whole inbox, even if it promises to summarize it for me. I barely trust Google with it.
Sierra: And it needed near-perfect reliability. The whole pitch was saving you time.
Marcus: That's the selling point.
Sierra: Exactly. And if you listen to your audio briefing, but then you're still like, beat, 'Wait, did I miss something? Do I need to double-check my inbox for meetings?' Then the core value, it's just... it's just beat broken.
Marcus: Totally. And if it gives you more anxiety, like, 'Did it really get everything?' then what's the point? It's supposed to give you peace of mind.
Sierra: And this last point, I thought, was super interesting: audio has a really low tolerance for error.
Marcus: What does that mean?
Sierra: Well, you know, you can skim mediocre writing, like, in seconds. Right? But with audio, you're just stuck listening to every robotic sentence, every awkward pause, every bad transition. It's just so much more jarring when it's right there, in your ear.
Marcus: Oh, that's a good point. Like, if a text email has a typo, you read past it. If an audio summary stumbles over a word, it pulls you right out of it. It's like a bad podcast, but it's your life they're messing up. (laughing)
Marcus: Speaking of things messing with your workflow, you know this other wild trend that's popping up with AI? They're calling it 'tokenmaxxing'.
Sierra: Tokenmaxxing? What even is that?
Marcus: So, basically, companies are starting to treat AI usage—like, how many tokens you're burning—as... what's the word? ...as proof of productivity.
Sierra: Okay.
Marcus: Yeah, it's creating this weird status game where employees are just burning tokens, right? Just to appear 'AI-active' without actually creating any real value.
Sierra: Wait, so people are just running AI queries for no reason to look busy?
Marcus: Exactly! The article mentions some Meta employees leaving AI agents running for hours, just to inflate usage numbers.
Sierra: Just running it?
Marcus: Yeah, like, one person apparently consumed two hundred eighty-one billion tokens in a month.
Sierra: What? That's insane.
Marcus: I know! Billions! beat And get this, Uber burned through its annual Claude Code budget in just four months because people were just, like, using it so much, but not necessarily for outcomes.
Sierra: That's wild. It's like the opposite of productivity, then. You're just generating noise to hit a metric.
Marcus: Yeah, it became about optimizing for visibility, not for actual output. Which makes sense, because it's hard to measure real AI productivity, right? But it's super easy to measure how many tokens you're using. I mean, they just—so they started tracking that.
Sierra: Okay, that's a totally different kind of risk, but still, like, an existential one for a company.
Marcus: Right? But back to Huxe. I still wonder, is it really impossible to compete with these big platforms? Like, maybe Huxe just failed on execution. If the product was truly ten times better, ten times more reliable, would people have switched?
Sierra: I mean, that's what we always hear, right? Build something ten times better. But then you run into these issues. The trust, the distribution. You're asking people to completely change their habits and give a new startup access to their most sensitive data. It's a huge ask, even for something amazing.
Marcus: But what if it was that good? Like, it never missed an email, the voice was perfect, it anticipated your needs. Wouldn't that be enough to pull people away from Google or Spotify?
Sierra: I don't know, Marcus. I think the gravity of these big platforms is just so strong. They're already in everyone's pocket, and they have all the data. Even if Huxe had the best product, how do you even get it to people when Apple can just push an update?
Marcus: So you're saying, no matter what, if you build a consumer AI app, you're just doomed to become a feature for someone bigger?
Sierra: I'm not saying doomed, no, but it certainly feels like you need a clear defense.
Marcus: A moat?
Sierra: Exactly! Like, is it a unique dataset nobody else has? A fanatical community? Or a niche so specific that the giants just won't even bother? Huxe, it sounds like, just... didn't have that.
Marcus: Man, it's a tough world out there for these AI startups. You build something cool, and you just become a case study for why not to build that thing.
Sierra: It really is a cautionary tale, for sure. I'm Sierra.
Marcus: And I'm Marcus.