According to The Verge, Sari Azout, the founder of the curation and ideas platform Sublime, is implementing AI deeply into her product while aiming to preserve its human-centric core. She detailed this approach on the second episode of a two-part Vergecast series focused on how developers are building AI into their products. Azout specifically highlighted Sublime’s new tool, Podcast Magic, which relies extensively on AI models. She also discussed using AI personally as both a productivity booster and a creative partner, expressing some reservations about future dependence on the technology but confidence in finding a workable balance.
The human in the loop
Here’s the thing: Azout’s perspective is a refreshing counter-narrative to the “AI is coming for your job” panic. Sublime is a platform built on taste—arguably one of the most human qualities. So the idea of pumping it full of algorithmic models seems, at first glance, contradictory. But that’s exactly what makes her take interesting. She’s not trying to offload creativity to the machine. She’s talking about using it as a collaborator to make connections a human might miss, or to handle the grunt work of sifting through endless information.
And look, this is the practical, near-term future of AI for most creative and knowledge work. It’s not about a single prompt generating a masterpiece. It’s about a messy, iterative back-and-forth. You throw an idea at the AI, it gives you ten variations, you hate nine of them, but the tenth sparks a new direction you wouldn’t have considered. That’s the “augmentation” model. The big question, though, is whether this just becomes a crutch. If you rely on the AI for those initial sparks, does your own spark-generating muscle atrophy?
The data and the doubt
Azout mentions the importance of good data, which is absolutely critical. An AI trained on the internet’s landfill is going to have terrible, generic taste. For a platform like Sublime to work, its models need to be refined on high-quality, human-curated inputs. That’s a huge advantage for a company that starts with a strong point of view. But it’s also a massive operational challenge. Curation at scale is hard. Can you maintain that quality of data as you grow?
She also expresses reservations, which I think is healthy. Everyone building with AI right now should have a nagging voice in the back of their head. We’re wiring this stuff into the core of our products and our daily lives without fully understanding its long-term effects on how we think, create, and even socialize. The confidence in finding a “balance” is optimistic. Basically, we’re all running a giant, real-time experiment on ourselves.
Podcast magic and productivity
The new Podcast Magic tool is a perfect case study. Podcasts are long, dense, and hard to search. Using AI to summarize, extract key points, or find specific moments? That’s a no-brainer utility. It’s using the machine to overcome a very human limitation: our time and attention. This is where AI truly shines as a tool—taking a cumbersome process and making it frictionless.
But when Azout talks about using AI as a “creative partner” in her own life, it gets fuzzier. What does that partnership look like? Is it brainstorming business ideas, or drafting emails? There’s a vast spectrum there. The risk is that we start to conflate productivity hacks with genuine creativity. Generating 50 social post ideas in 10 seconds is productive, sure. But is it creative? I’m not so convinced. The real test for Sublime will be whether its AI-augmented curation feels inspired and surprising, or just efficiently predictable.
