No AI tool puts the fear of God in me like NotebookLM.
I feel an existential, jealous dread whenever I see other people using it. Turning notes into podcasts. Ideas into slides.
As anyone who has worked with Agents can likely tell you, it still takes a lot of tedious human judgment to make AI output feel right. But with NotebookLM, something that once belonged to a small group has seemingly leaked into the open.
Sofía López articulated this growing dread so perfectly in I work in AI and I’m scared. “I’m scared of something pathetic and real: that I’ll waste this moment.”
Andy Coenen also explored the role of taste in isometric nyc: “The quality of art is defined by the quality of your decisions - how much work you put into something is just a proxy for how much you care and how much you have to say.”
I want to take these ideas and play them forward. There will always be a place for human-created work. Possibly an even more revered one, as AI scales. But AI isn’t coming for the sacred or the sublime. It’s coming for the useful.
I think NotebookLM is a prototype for a new economic model. It proves that when execution becomes trivial, the value shifts entirely to the format. This piece is my attempt to map that shift. NotebookLM is the first real glimpse of format-as-a-service at scale. FaaS?
It signals a future where creators won’t sell labor or output, but will franchise their judgment through APIs. Format becomes the product, and taste becomes the only moat left to scale.
We all love NotebookLM. This is bad news for NotebookLM.
NotebookLM is a research tool. You upload documents, it spins up a small RAG pipeline you can interrogate. It’s a really great product.
But what users actually get excited about are the outputs. Podcasts, summaries, study guides, slides. The joy comes from the shape the information takes and the godlike ability to summon this content at will.
NotebookLM is a winning formula because someone at Google obsessed over the transformation pipeline. In other words, the specific way raw information gets restructured into a format that feels good to consume.
Here’s an example of an automated Podcast I created from Sofía and Andy’s articles. I even got that same small creative rush from listening it back after it was created. But this podcast is a product that tens of thousands of others can build near-instantly. Once you realize that, the perceived value of its output decreases immensely. A NotebookLM podcast is miraculous the first time you hear it. By the tenth time, it’s just noise. So where will this arms race end?
To stand out in a world of infinite generation, you don’t need better AI. You need novel formats. Businesses will want a format that feels unique. And because novelty fades fast, they might use a format once and then discard it.
That’s okay for NotebookLM, though, because very soon we’ll be able to change our podcast hosts. Or even make ourselves the host. We’ll all have a podcast. At last.
With NotebookLM, Google has done the tedious work of understanding how and why a talk show format is successful, and have encoded that taste into a pipeline. They’ve also done it at scale. The AI does the execution, but the format itself is the product. In our new ‘vibe economy’, format is the vehicle for value.
Format-and-expertise-as-a-service
We’ve seen some of this play out already. We’re now witnessing the rapid evolution of skills.md packages — bundles of markdown documents that tell the AI how to behave. “Don’t use the word leverage.” “Call this MCP tool.” What I’m describing is probably further down the road: taste encoded into real pipelines.
What’s missing in this future is a marketplace for prebuilt, high-taste content formats. These formats could look like something we already know and love:
WIRED Question / Answer Videos but for your C-Suite leadership
Every Frame a Painting but for your product explainer videos
This American Life but for your SDR emails
But also it could look like:
A video case study generated for every customer in your business
A competitive battlecard produced dynamically as the prospect states objections live on a call
A customer persona document built with real Gong recordings
When a company hires a strategist today, they’re paying for three things bundled together: the thinking, the structure, and the execution. The strategist figures out what to say, decides how to frame it, and produces the deliverable. Format-as-a-service unbundles this.
The company brings the thinking: their data, their context, their specific problem. The format provides the structure: the decisions about what goes where, what gets emphasized, what gets cut.
This is a strange inversion. We’re used to paying for labor. Now we’ll pay for the shape that labor used to produce, divorced from the labor itself. If you’re a creator, the biggest opportunity I see ahead is in franchising and packaging your formats in a way that will allow them to scale beyond yourself. Companies will pay for this.
So where does this lead?
We’re about to see an explosion of AI-generated content. Yes, most of it will be beige. But some of it will be genuinely good. Not because the AI is better, but because someone with taste designed the format it flows through.
Most creators will keep selling their time. A few will figure out that their job isn’t making things anymore. It’s designing the framework that things flow through.
And when they do, they won’t sell content. They’ll license the shape.