"Descript Isn't a Slop Machine": Laura Burkhauser on the AI Tools Creators Love and Hate

Today my guest is Laura Burkhauser, CEO of the pioneering video-editing platform Descript, which originally burst onto the scene in 2017 with its revolutionary AI-powered, word-processor-like editing paradigm and, as you'll hear, has continued to push the boundaries of what AI can do for creators ever since.

Laura took over for Descript founder Andrew Mason, who was my guest on the show back in August 2024, after serving as VP of product for several years. As a longtime Descript customer and early adopter of its new Underlord API, I've been impressed both by the company's customer obsession and its product velocity. So I was genuinely excited to get Laura's take on product management in the AI era.

We begin with a remarkable email that Laura recently sent to customers, in which she recognized that generative AI is a polarizing topic among creators and declared, quote, “Descript isn't a slop machine, and we don't want it to be.”

For me, this begged the question: What is slop? For Laura, who emphasizes that all creators have to start somewhere and that all new media takes time to mature, it's less about the quality of the content and more about the incentives that drive its creation. In short, it's the mass production of content for the explicit purpose of algorithmic attention arbitrage that she objects to.

In this, she's in step with Descript's creator customer base, who she says approach AI with a passionate mix of enthusiasm and hostility. Narrowly scoped, purpose-built, and critically reliable AI tools, such as Descript's Studio Sound, Green Screen, and audio Overdub features, are pretty much universally loved.

Underlord, their natural-language-instructable AI editing assistant, which I personally do find quite useful, is something everyone wants to love, but many still find frustratingly limited. And then there are the infamously unruly image- and video-generation models, which, despite—and perhaps in part because of—their soaring popularity, are the object of visceral hatred.

It's a lot to manage, particularly with general-purpose products like Claude Code accelerating to the point where they're starting to become capable of video editing. But Laura's true north is simple: It's her job to make sure that, no matter how good frontier models get, you have a better experience using Descript than you would with an AI agent alone.

To this end, we get Laura's razor-sharp takes on how Descript decides which generative models to include in the product; why they plan to use frontier models to power agentic editing for the foreseeable future, while also training task-specific models in-house where they happen to have a unique proprietary data advantage; the critical importance of, and challenges associated with, multimodal understanding; the critical role that expert aesthetic judgment plays in the process of model evaluation and iteration; and the product-design principle that says AI assistants should be able to do everything that human users can, and vice versa.

We also discuss how Descript is designing the Underlord API to be hired by coding agents, as well as the pricing and design challenges that arise when a single button click or API call can consume multiple dollars' worth of credits.

Finally, we take stock of where we are in the big picture. Laura emphasizes that while economic logic might dictate a future of infinite slop, artists have a long history of adapting to and incorporating new technologies in unpredictable and often defiant ways. She's betting that our cultural reality will be far more vibrant than our Black Mirror fears.

Hi, Nathan.

Thanks for being here. I'm excited for this conversation. As longtime listeners know, we are Descript customers and use Descript to help produce the podcast. This is actually the second Descript CEO episode on The Cognitive Revolution, although the person holding the seat has changed. You're new in the role.

That's right.

I looked back at a couple of emails that you sent, including one that you sent, I think, right after taking over. You wrote something that I think will be a really great jumping-off point for us: “Descript isn't a slot machine, and we don't want it to be. So how do we keep building generative AI features without surrendering to the slop? Or is it impossible?”

That's going to be a defining question, honestly, in terms of how people, even in the big picture, spend their time over the next few years. So I wanted to start with the first question that comes to mind for me: What is slop? How do you know it when you see it?

I might define it differently than other people. To me, I think about slop as being a form of content arbitrage. It's when you can identify a temporary, almost an inefficiency or opportunity in the market to create content that is likely to give you a return on your investment.

In this case, it's that you can pump the system with a lot of content that is extremely cheap for you to make. That might not get a ton of engagement or a ton of subscribers, but it gets you enough revenue or engagement from that content that it ends up being net positive for you.

There are people who can identify these kinds of slop-arbitrage moments and really take advantage of them. The 2 key elements of slop, to me, are that the incentive is money in some way, ultimately, and that it is happening at scale.

I think there's a lot of bad art out there, and I would say that generally I'm pro-bad-art. I think bad art is a really important stage that you have to go through to get to good art or good content.

I don't know if you can remember the first few things you put on the internet, Nathan, but my guess is that you would cringe if you looked at them now, knowing how sophisticated you've become in your creations. To me, there's a difference between slop—which is, “I'm trying to pump, I'm trying to juice the algorithm, I'm trying to fill YouTube with a bunch of avatar meditation videos in this moment when it hasn't caught on so that I can get some ad revenue real quick”—and thinking, “Maybe I should be a meditation guru on YouTube. I'm going to create this avatar.”

The result may be the same, but I don't think that's slop. That's just someone's bad idea.

Yeah, interesting. So if you're feeding the algorithmic hogs, you are producing slop. I wonder, though—and by the way, if the comment section is to be believed, I'm still in the bad-art phase of my own personal development. We'll see if that ever comes to an end.

One thing I will say about AI is that it is allowing me to create stuff that I don't think is terrible, at least, and that I enjoy the process of creating in ways that I just never would have had any opportunity to do before.

For now, I'm ignoring the haters in the comments, who are almost uniformly opposed to my new AI-generated YouTube preview art. I'm having fun doing it, and I kind of like the look of it. So for now, we'll keep going.

What I mean, though, is that's exactly what I mean. If you were learning how to paint for the first time and you sat in front of a canvas and painted me a picture, it would probably be really bad. You haven't figured out what your voice is, what your aesthetic is, or what feels good to you. You haven't taken any classes. You haven't played with the paint. That's your very first thing.

This is a new medium. A lot of the “we're having fun with a new medium” is how you get to good stuff, right? I'm sure that you're finding that as you play with this stuff more and more, you're starting to have your own opinions: “I don't like this,” or “This prompt is working better. I'm liking what I'm getting more.”

You're going out into the world, seeing other people who are doing stuff, and thinking, “I like that person's style. How can I show up in that same way that that person does?” That's the same way that, if you were learning how to paint, you would be developing your painting style.

I'm also not a hater who thinks anything created with generative AI is slop. I think that, because this is a new technology, most of us are in our “create a lot of bad stuff” kind of phase. That's also because the technology is still nascent.

But I am very bullish that it will be possible—and already is—to create really awesome stuff with generative AI, stuff that is not slop at all. I think the only way you're ever going to get there is by creating a lot of bad stuff first, because that's how you get good at everything. That's how you get good at art.

There's no doubt in my mind that good-quality stuff can be produced with the new generative AI tools. My creative teammates at Waymark are clearly head and shoulders above me in terms of their ability to do that, and it's a little bit hard sometimes to put our finger on exactly why they're so much better. But I think it's pretty undeniable that there is still a significant skill gradient in terms of what people can do.

They've also put in the time to a much greater extent than I have, even still. And I say that as an early adopter and enthusiast of just about everything AI.

Yeah, and they're also fighting their tools right now. I think there are 2 things that are really blocking us from seeing a lot more good generative AI art or content, or whatever you want to call it.

And the first is that the technology is just not quite there yet. It’s getting better, but no one would actually choose to make a video by generating 5 to 10 seconds of it at a time and crossing their fingers that the voice consistency between clip 1 and clip 2 is good enough that no one notices, or deciding they’re just not going to have voice in video. Right now, there are all kinds of constraints. You have to generate something like 50 times sometimes to get it exactly the way that you want it.

So if you’re making good GenAI stuff right now, it’s because you’re super invested in the medium and you want to fight your tools the entire time. Then I think the other reason we’re not seeing a lot of it is because there’s a lot of stigma right now for people with the kinds of taste and skills to be able to make good stuff. There’s a lot of stigma for them to be using these tools, publishing, and owning that they’re using them.

And so when you go on X or wherever it is that you’re consuming social media, a lot of the people that you’re seeing using these tools are people who are earlier in their journey of knowing what good looks like visually. So it is bad. I think the average person who’s never studied—whether by “studied” I mean in school or through experience—film or photographic composition is going to need to spend some time in the medium before the stuff they make is actually good, and they’re not going to know why.

You’re going to be like, “I know this isn’t good. I don’t know why. I don’t know if the calculators say why.” If you stick with it, it might get better. It doesn’t for everyone. But those are the 2 reasons: right now, we don’t have the tastemakers really using it, and right now they really have to fight the technology for it to feel good, for them to ever get into a flow state, for it to really feel fun for them. So you just have early adopters that are really digging into this stuff right now.

That comment on vocabulary, I think, is a very apt one. I had just said it’s a little hard to put our fingers sometimes on why the creative team is so much better than me, but I think that is a huge part of it. So often, if I send our creative lead something that I’m working on, he’ll give me a few adjectives or an artist’s name for inspiration, something like that. In some cases, it could even be an explicit direction for composition, and that really does usually take it up a pretty clear notch. So I think that’s a great detail to highlight.

I don’t know if it was in the same email or another one, but you described adding generative AI models to the Descript product as a polarizing topic, and I guess a polarizing product move. I’ve been broadly very impressed, by the way, with my interactions with you. We’re on the API early-adopter beta list, and I think your team is very plugged into what people are trying to do and how you can help them do it. So I’ve been very impressed by my interactions with the people building the product.

Who signed up for this? What did they want to tell you? What are you hearing from people, and how has your sense of what is making it polarizing—and how the discourse around these tools is evolving—changed since you formed that committee?

It has been fascinating. First of all, there was a huge uptake in that invitation. We got way more people who signed up for it than we could convene in a reasonable way, although I think I’ve talked to most folks who signed up for it now.

Why did I send this email in the first place? When I became the CEO of Descript, I had been the VP of Product for several years before that, so I wasn’t brand new to the discourse. But we had just changed our pricing, which was a tough moment. We’re a really popular product with a huge, loyal, and vocal set of customers that we love dearly, and I’m glad that came through in the way that we’ve talked to you. We are obsessed with our customers, but we did need to change our pricing, which is always kind of a moment, and we got a lot of feedback on it.

There was one type of feedback that I thought was really interesting and wanted to dig in on. It came up a bunch in some form of, “I wish you would stop spending time building AI features and use that time to invest in the core quality of the app.” When you see feedback like that a number of times as you’re raising prices or changing prices, you should get really curious, because there’s a lot that could be in it.

There are a couple of things to unpack. The first is, do we have a problem with core quality, and what do you mean by core quality? What is core quality to you? Are we talking about performance? Are we talking about reliability? Are we talking about upload speed? Are we talking about playback speed?

Interestingly, there were some really good things that came out of that, and I hope folks see that we’ve been knocking through a bunch of quality things. But then there was also a bunch of core quality being new features that, in fact, to my mind, are AI features. One of the things that became really clear to me is that when we talk about AI features, different people really mean different things.

Descript is an AI-native product. We’ve been AI-native—the whole idea of editing a video like a transcript is actually an AI idea. The way that we implement green screen is by using an AI model, a visual model. The way that we do Studio Sound is with an AI model that we created. We do voice cloning and generate an Overdub, so that you can change something when you realize, “Oh, I recorded the wrong word. Let’s just go back and change that,” and have it in your voice say the right thing. Now it lip-syncs, and we built that model. These are all AI products.

Many of the features they wanted us to improve are actually AI features. So what I came to understand is that there’s a hierarchy of hostility toward different types of AI features. Descript users, at least, love a lot of AI features, especially when those features are effects, transitions, things that have a button that does something that feels deterministic to the video, even if it’s powered by AI. Green check mark—everyone loves it. Keep building that into infinity.

Then there’s Underlord, which has our AI co-editor. Underlord is somewhat polarizing. Everyone wants it. People are very excited about speeding up their workflows—their AI editing workflows—with AI, and they love the idea of an agentic co-editor that helps them do that. But they’re mad that it’s not as good as they want it to be for some of their use cases.

So it’s not “Don’t build Underlord”; it’s “Why isn’t it perfect yet?” But I do want this. I do want you to build something that helps me get through the drudgery of editing faster. That’s the general sentiment on agentic co-editing.

Then, whence all the hatred? It is really generative video that’s the polarizing topic, to some extent. Avatars, not so much. Voice cloning and TTS, people feel pretty good about that, but it’s really generative video.

When I dig into it, I think—and we’ve talked a little bit about this, so I don’t want to belabor it—there were really 2 things that people didn’t like that made them mad about this, in addition to some of the other general “pause AI, stop AI” kind of stuff that’s in the air. Strictly speaking, from a creator perspective, I think there’s a feeling of, “I feel like I’m going insane because everyone is telling me that this stuff is super good, and it sucks, and I hate working with it.”

I think that’s a very reasonable perspective for the average creator to have because, as I said, right now, to have a good experience, you’ve just got to be really invested in the medium because the technology is hard to use. We don’t talk about that enough in all the hype cycle. People feel like they’re sold this story that this stuff is amazing and incredible and the future, and they use it and they’re like, “What’s wrong with the world? I feel like I’m taking crazy pills here.”

That’s one part. Then I think there’s this idea that goes along with the hype-cycle part of this discourse: How many times have you seen someone say something like, “Sora just put a gun to Hollywood’s head and pulled the trigger”?

It’s like, yeah, okay. If that’s how you’re going to sell your technology, the people who you just said got a gun put to their head aren’t really going to like or be excited to use the technology. That’s not how I talk about this stuff. That’s not my perspective—that this stuff is going to end the role of traditional film, that it’s going to end recorded media, or that it’s going to put everyone who works in traditional media out of a job.

Personally, I still believe that. I think this is a new tool, like many new creation tools that we’ve gotten over the years. Film was a new tool, and this is generally pretty exciting. I don’t know. Is it threatening? I don’t know that the main story should be that this is threatening and going to displace a ton of jobs, because it’s not clear to me that it’s going to displace a lot of jobs. I think it may also create a ton of jobs. It may shift jobs.

But this is simply a creative tool that we ought to be approaching with fun, a sense of play, and curiosity. Instead, because of the discourse around it, it is perceived as threatening and overhyped.

I love the emphasis on play. That's one of my most common refrains as well. This technology rewards play, and not just the video or visual generative models, but really all of the current frontier AI capabilities reward play more than any other technology I've ever used. That really is the right mindset to go into it, and I couldn't agree more with that.

I think there are 6 different follow-up questions that I want to ask based on everything that you just told me. Maybe the first one would be: How do you choose which generative models to put into a product? There are obviously many. They have very different strengths and weaknesses, and they have different price points. There's another question about price and how you're thinking about and managing that.

But these things are super hard to benchmark, right? It's not like, in the Underlord portion, I think you'll have a much clearer line of sight to whether a new model or new prompt or whatever is doing what we want it to do in a reliable way for a finite set of understood use cases. With the generative stuff, it's tough. Is it just vibes, or do you have a better answer for how you're figuring out what to actually pull the trigger on and move into the product?

There are 2 stage gates. The first is, should this be available within Descript? The second is, should we make this the default model? Because most people are not going to change the default model. They're going to accept whatever you put as the default model, right?

That actually might be surprising, too. I feel like that's something that, if you're deep in AI, you're sort of like, “Why aren't you using the model picker? Obviously, Nano Banana Pro is going to be the best thing for photorealistic face swaps, but then you should be using Kling for this other usage, right?” That's how people who are deep in AI think about things, but the average person doesn't have that level of sophistication and doesn't want that level of sophistication.

How do we make decisions about default models? How do we make decisions about what models to improve or bring in? It's a little bit vibes. I'm not going to lie, because it's not like we evaluate every single model out there and say, “These are the 5 best,” or whatever.

Often, it needs to be available through Vowel, which we use as our provider. If you're not in Vowel, you're not going to be in Descript, because we don't want to build our own custom connector for your thing unless it's the best thing ever. That means we need to sign a new data license agreement and all this stuff that's like, “What a headache.” We've already done it with Vowel, so we're just going to do it there.

That's why C Dance is now in Descript: It is finally in Vellum, so we're like, “Great, you can come on in.”

Within the stuff that's in Vellum, we try to pick the stuff that feels like it's generally the best in the game. What you see are these standard industry benchmarks of these different things, and you'll see that you have the same labs on the leaderboard month after month. We try to make sure that we have some representation from each of those labs, because you're always 1 week away from that lab coming back to the top and having the best thing available.

When it comes to the default, that is where we do look at external evals. Then we run some of our own on common customer use cases to find out where we generally think people are going to have the best experience.

For image generation, Nano Banana Pro, I think, is our new default. What we then do is A/B test it against the existing default and make sure that we're seeing good things from the A/B test, and that the A/B test matches what our internal evals tell us. If it does, then it's a definite ship: This is our new default.

When you do an internal eval, is it a panel of trusted people who are scoring outputs?

Yeah, it is.

Yeah, interesting. That's a hard thing to automate. We have done some of that stuff. It's been a minute since I last did a version of that.

I was also struck that our original use case was a little bit different. So with Waymark, we have this TV commercial maker for small businesses. For a long time, we've had a tool that pulls in all the images that the small business has published to its website or Facebook page or whatever, and then we make this library.

That was a great convenience factor even 5 years ago, when there wasn't much we could do with it beyond saying, “Here's what we collected for you, so you don't have to collect it for yourself and upload it.” But obviously, with AI, there's a lot more we can do.

The aesthetic quality of an image was always a really hard thing to evaluate. In the early days, you could caption it, but did it look good? Models had a hard time with that.

Absolutely. I'm not even sure that you should be trying to automate that. I went to this dinner with a CEO of Midjourney, and he's like, “The reason why we still have the best, aesthetically, image generation is because I have my thumb on the scale.”

Google just lets some kind of democratic panel or automation decide what the best image is, and the best image is always some generic pretty blonde lady or whatever when you ask for something. I thought that was pretty funny.

But all of that is to say that this may be an unpopular opinion, so it's fun for you because it's a little controversial, but I don't think you should underestimate the importance of vibes in aesthetic evals.

When we first built Studio Sound—which, by the way, is still our internal model—we recently evaluated it against all the other new Studio Sound providers, and we still chose ours, even though in other cases we've thrown ours out and taken another model that's obviously better. We kept Studio Sound.

Studio Sound was originally built by a cellist who just had a really good ear for things, and he did what we might now be calling an eval. Our original eval process was this guy would listen to different models and be like, “This is better. This is better.”

When he left Descript, we had to actually write an eval that was like, “What are the 37 different things that make one form of background-noise removal better than another form of background-noise removal?”

I don't know. I think it's reasonable to say that, in something that is primarily judgment-based, we're just going to have a human do this, and we're going to have a human do this forever. Someone we know and have vetted as having good taste is going to make these decisions.

Yeah, okay, that's quite interesting. Could you do just a quick overview of the frontier model landscape as you see it? You started to a minute ago when you said Nano Banana is best for face swaps and then Kling does the other thing.

Is there an expanded version of that where you would say, “Here’s how users should generally orient themselves to their options”?

You mean specifically for video and image generation?

Yes. Or others, if you have a similar account for them. But yeah, that’s what I was thinking.

I’m probably not the best person to ask about this. Do I have a cheat sheet here that tells me that? Honestly, I’m probably not the best person to ask about this. But I know that we generally have a perspective about it. I know that our defaults right now are Nano Banana Pro and Veo from Google, and that we’re considering replacing Veo with C Dance. What’s interesting is—I’m not going to get into that, actually.

I would say that, especially when it comes to video generation, I don’t think there’s going to be a winner-take-all across all of these generative image and video models. The use cases for generative video, for example, are so different that it’s very difficult for me to believe that the same model is going to be the winner for something like Oscar-film-worthy special effects and making the cheapest but high-quality-enough video for all of the product pages on Amazon.com.

I just think there are going to be models that are really good for massive bulk actions that don’t require things like consistency across time, sound, or voice. Those things aren’t going to be important. It’s going to be about a bulk play versus something where, if the quality is really good, you’ll pay thousands of dollars for generation.

Like most products, I think it really helps to understand who your core customer and hero use cases are. One of the things that came up for me with C Dance is that it’s an amazing generative-video solution if you want a really opinionated edit where it’s going to make a lot of artistic choices for you that you didn’t necessarily ask for. If you’re someone who’s wanting to abdicate a lot of that, or who’s able to describe exactly, ahead of time, beat by beat, what you want to happen, you’ll have a good time with C Dance.

A lot of people at Descript are using generative media as B-roll. C Dance then feels almost too flashy and directed when you use a general prompt, so it can be distracting. B-roll is generally supposed to be not super distracting and not take too much of your attention. Your attention should be on the A-roll.

For the typical use case, which is not using it as A-roll but actually using it as filler B-roll, I don’t know that Seedance is the right choice. First of all, C Dance may be overkill. It’s too expensive for that use case, and it may be too much of a scene-stealer for that use case. You told me it was okay to get in the weeds.

Yeah, please.

That’s an example of getting in the weeds. I rely on people to help me understand these different frontiers more and more all the time. I used to be able to try all the new models myself 3 years ago, and now it’s just getting to the point where I have to rely on the network to help me understand them. So, please don’t shy away from any of the nitty-gritty detail.

Okay, cool. I think that’s really interesting stuff on generative models. The perspective that there won’t just be one winner makes a lot of sense, too.

I actually think that’s why a lot of the orchestrator agents are going to need to be good at understanding which generative model, or which other kind of model, they’re going to need to orchestrate between all of the different models. They’ll need to understand, given the context they have about the video or the project that the user is working on, which model is likely the right one to use, and why, or whatever. It kind of hits their cost, quality, and use-case bull’s-eye.

Makes sense. Well, let’s get into Underlord, because that is obviously the agentic interface for Descript these days. One thing I would love to understand, again, in as much detail as you’re willing to share, is how the AI—or AIs, as the case may be—sees the video. How does it understand video?

I’ve done a bunch of stuff with this over time, even with Gemini, which is video-native in some sense, in that I can throw a video at the API and it will accept a video file. I’m not quite sure what’s going on under the hood. Are they taking frames out of the video and doing some sort of sampling?

It doesn’t always feel to me like it’s truly watching the video. Sometimes I’ve asked it to critique videos, and it sort of says that there are hard cuts when there weren’t hard cuts, just because I moved my head between 2 frames or things like that.

Obviously, video is a huge, heavy file in the first place, right? A big part of video software over time has been managing that, and now we’ve got another generation of that problem as we provide video as inputs to models. How does it get processed, and how does it get structured so that it can be presented to 1 or more AIs in the most effective way?

Right now, we translate visuals to text and consume that. We do what’s called captioning, so we do frame-by-frame captioning of what is in each frame. Then we use some clever tricks to fake giving the agent eyes and ears that way.

I think it does okay. I think this is an area of huge opportunity for us, and working multimodally is right now the agent quality team’s number 1 priority. I would stay tuned here to see a major upgrade in the next month or 2. But right now, we do visual captioning.

This also connects to the idea of training your own models versus going out and getting models. It sounds like, if I interpret your previous statement correctly, you’re kind of neutral on whether it’s your own model or somebody else’s model, and really just focus on what’s going to deliver the best, maybe cost-adjusted, user experience. How do you think about choosing between build versus buy when it comes to models?

We have a strategic bull’s-eye of where we aspire to have the best models. I’d say that where Descript aspires to have the best models is when you start with recorded media and you’re editing recorded media. We want to be the world’s best at that job.

That’s something we call Regenerate. We can go into the recording that we’re in and, a few questions back, I may have said, “I really don’t like my answer. Can you actually make me say this instead?” You can change my voice and my lips to say the thing that I wish I had said in the first place.

That’s a great example of 98% recorded media, but we need to update all of our branding to say this, or we need to update the dates, or California just added a new law and we need to change some of the language from the way that Laura explained this concept 2 years ago. Can we just regenerate that without having to re-record the whole video? That’s the kind of job Descript wants to be really good at.

We’re about to launch smoothing jump cuts. If you do have some crazy jump cuts, you can use Descript, and we’ll just make it look like you naturally moved over there and never made the edit in the first place. Those are the kinds of things where we have Descript models to do that, and we want to own that space.

For purely generative stuff, we’ve said that we don’t want to own that space. It’s very expensive to build those models, and I think most of the companies that are spending hundreds of millions of dollars to build those models are still going to lose to Google. I don’t want to set money on fire that way.

I think it’s about deciding where you want to win and then deciding where you can borrow. We are very friendly to borrowing, especially around pure-generation kinds of things.

I think I’m really excited about this, but I think this is where you get into a blurry line: heavily augmented recorded media. For what it’s worth—you didn’t ask, but I will say that there is a part of me, maybe a stodgy part of me, that just feels possessive about human expression and human facial expression, and feels a little unsettled by AI clones, or something that purports to be me but isn’t.

At the same time, I’m very sympathetic to the idea that I had to turn on a whole bunch of lights in this studio, and I had to make sure that my makeup was decent before I came on to record with you. Wouldn’t it just be nice if we could have this conversation in an authentic and human way, where we don’t type on a piece of paper and then add my voice to it, and then add my robot face to it?

We mostly just talk as humans, but then in post we could do all kinds of magic to make it look like I was wearing makeup and looked amazing, had a great outfit on, and the light was perfect, and I didn’t say anything stupid. So, that’s the vision that I have for Descript: to really make the killer use case that we’re better than anyone at—augmented human-recorded media. That’s where we really like to build.

Yeah, cool. That’s quite interesting. I guess I’m sort of feeling out the strategic landscape of models. It sounds like some of the models that you’re building may have no offerings on the market. I’ve not seen one, for example, that does jump-cut smoothing, and I could also imagine—you’ve got retake removal now, and that’s been there for a while—but I’m also guilty, all too often, of vocalized pauses.

So, we can remove ums and uhs and that kind of stuff in a pretty smooth way. It strikes me that the structured nature of the edits that people make in Descript is an unbelievable data set for some of these use cases that probably nobody else has even collected the data on. I feel like I’m intuiting where the core advantage lies, based on all the work that people have done in the product over time.

But maybe you could tell us a little bit more about how you think about the data flywheel. If I was going to say what I want Overleaf to be better at, it would be some of these subtle things where I stuttered, repeated myself, or whatever. Which one do I cut? Do I cut the first version that I said, or do I cut the second version that I said?

It’s not always the case, but I think you usually cut the first one. If you felt the need to say it again, then probably the second one would be better. Often, when I highlight a word and hit ignore on it, I go back and watch that passage again to see how it landed—whether it was glitchy in any weird way or whatever.

I think if there were a model that could make those marginal decisions well, where you try this edit, try that edit, and see which one looks better, that would be an amazing upgrade. That right now is the bulk of the time that I spend in Descript that I would love to offload. I just made that edit—how does it look? Let me try it with the alternate version—how does it look?

Making good decisions there would be an amazing upgrade. It doesn’t sound like that’s something Google is going to solve anytime soon, or anybody else. I don’t know; you have other candidates out there, but it sounds like you’re probably going to have to do it at home.

That’s right. That’s exactly right. We tend to invest where we think we have great data, where we think we can build something without breaking the bank, and where we think it’s unlikely that one of the labs is suddenly going to care a ton about removing retakes. That’s much less interesting to Google, I think, than solving voice consistency between 3-second clips or whatever.

So, yeah, you’re intuiting correctly. If that’s where you want us to work, first of all, I’ll pass your feedback on to the product team. We’re taking things on chunk by chunk, but that’s where the thumbs-up and thumbs-down really helps us identify where we’re not hitting the quality bar, either in an AI action—which is one of the deterministic-seeming tools, like Remove Retakes—or in an Underlord request itself.

You might say something like, “Get rid of all of my filler words unless you can’t make a clean cut. If you can’t make a clean cut, keep the filler word.”

I haven’t tried giving it the freedom to determine that it can’t make a clean cut. Is that something I should be doing more of?

Yeah, you should. We’ve tried to make Underlord truly open-world. Underlord doesn’t just have 20 tools that it can use, so if you choose something like idea 29, it doesn’t know what to do. It ought to be able to handle nearly any request, though not all with equal ability.

We use the thumbs-up and thumbs-down to understand where the pain is. We have an eval set, which I can tell you about, but then we use user feedback to help us understand where the pain is, because there are all kinds of things that we’re bad at. I’m just going to be real: there are some things we’re great at.

In our evals, we have 3 grades that you can get for a user request. The first is that you didn’t break anything. You just didn’t break my video. You didn’t do something that made me say, “Oh my God, you ruined everything.” That’s grade 1.

Grade 2 is that you did what I asked. I said, “Remove filler words,” and you removed filler words. Thanks, buddy. Doing it well would mean that you removed filler words and didn’t have these really striking jump cuts or changes in tone as a result.

The way that we do evals is that we take a random selection of real user queries and run Underlord against those in version 1 and version 2. Then we have a ton of LLM judges go through and grade them multiple times, and we take the average of those. We say, “Okay, this is the percentage where we didn’t break things.”

We aspire for that percentage to be close to 100%. We never want to break your stuff. Then there’s “did what I ask,” and right now I think we’re aiming for that to be 90% of the time: we do the thing that you ask us to do.

Then there’s “do it well.” Right now, with “do it well,” we’re okay. We could be better. I would like that number to be 80% by the end of the year, because right now we’re still not doing it really well. I think when you really feel like 80% of the time that you ask us to do something, we do it at about the level that you would do it, that’s when you really trust your AI co-editor.

That’s across all use cases. There are some nice spots where we’re doing it well a lot, and rough cut is an example of that. Whenever you’re asking it to help you with a rough cut, to help you get a long story into a shorter form, we tend to do that well a lot.

A lot of the visual stuff we’re not doing as well, and that’s why multimodal is a real priority for us right now. But user feedback helps us understand, “Wait, there’s a hotspot here. There’s a hotspot around filler words where people are just not happy. Why don’t we go spend a couple of sprints getting this part of the product really cleaned up?”

As you try to push the frontier on this, I can imagine a couple of different strategic directions you might go in terms of how to get the best performance out of the available models, with the various constraints that they have. Maybe you’re even doing multiple of these.

One angle would be to say, “Okay, Claude, GPT, or maybe Gemini is probably going to be the best reasoning and tool-use agent. So, what we really need to do is set whichever one of those we’re using up for success.” How do we do that? We need to give it a richer understanding of what it’s working with.

Since it can’t natively detect awkward moments, for example, maybe we need an awkward-moment detector that we can run and then feed into the model to flag when these things are happening, so that it knows to reason appropriately about that.

But then you can imagine a different version where you say, “I’ve been hearing very good things about GLM-4.5, and I think the weights are out there for this.” Maybe we want to try to do something deeper, where we actually teach the core model to understand some of these inputs.

Adding video as a modality to a GLM-4.5 doesn’t sound easy at all, but you could do some sort of late fusion, cross-training, or what have you. I guess this sort of decision probably depends a lot on what resources you have.

Do you feel like you can hire the team to do frontier work at that level, or is it so hard to compete with the frontier labs for that kind of talent that it’s out of range? It might also depend on whether we think open-source models are going to continue to be competitive, or whether we feel like Claude 5 is going to run away from the open-source models for compute reasons, constitutional reasons, or whatever else.

If it runs away from the open-source base models, then we can’t really keep up, even if we do get good at doing more advanced stuff on open-source bases. So, to bottom-line all that: what’s the model strategy? How do you think about what trends you want to bet on carrying you forward?

I think the main bet that we’ve made with our agent is to try to build a very generalized harness and give the agent access to a bunch of low-level tools, assuming that generalized intelligence is going to get better and better. We’ll use probably a handful of whatever models Anthropic, OpenAI, or Gemini comes out with, so that when the next Claude model drops, we’ll have it evaluated within 15 minutes and put it in the product.

Building for that, at least for the short to medium term, is the right bet to make, rather than investing a lot of time and money in research trying to keep up with the labs.

So then it's about: How do we build an agent harness that's going to be able to instantly take advantage of leaps in general intelligence? How do we not get bitter-lesson'd into not being able to immediately take advantage of those leaps? That's how we've tried to build our agent: give it a ton of context about the Descript model, about video editing, and about how to think about user requests, and give it access to our low-level tools.

And then we do some stuff. We have various experiments that we're doing to make it better through personalization over time.

Okay, I'm very interested in the personalization. Does that boil down to saying that you want Underlord to be essentially in the same position as your human users? Obviously, it can't see as well, and it can't hear in the same native way, but subject to the constraints of some of these things having to be arms-length tool calls to do the sensing, it sounds like, aside from that, you're building one harness for both humans and AIs at the same time. Is that a reasonable way to think about it?

I have not thought about it that way, but I think that is right. We do have a design principle that Underlord should not be able to do anything in the editor that a human can't do, and vice versa. They should all have access to the same tools.

That also aligns with the general design principle we have that Underlord is a collaborator with you in the editor, the same way that we've been a video collaboration tool since day one. We've been a tool that teams use because often video is not a solo job. It's a job that you do with a team, and Underlord is a member of that team.

Yeah, okay. That's quite interesting. I feel like I'm still seeing this in a fuzzy way, but it seems like an increasingly universal design pattern, which Descript doesn't quite follow yet, although what you articulate is very consistent with it. It's an app that you could use using your own intelligence: click all the buttons and do all the things, and then frame that with an agentic companion that, to varying degrees, you could say, "Do this one thing for me," or, "Do everything for me."

It would log for you what it's doing, using the same exact tools that you could use. In creating that log, you could also pretty easily go in and undo the one thing that it did that you didn't want it to do, that didn't work well, or whatever.

Do you see that form factor? Right now, there's not this very persistent Underlord presence that's a long-running agent where I can see everything it did. It feels like it's more embedded into the product as opposed to framing it and sitting outside the product. Do you think that's something that will change over time, or do you feel like maybe I'm going in the wrong direction with my framing?

I think that it will change over time, and I think we need to decide exactly how. Ultimately, Underlord would be more powerful if it had—right now, it lives at the project level. It needs to at least live at the drive level, but I think it would be even more powerful if it could live outside of the drive and be a collaborator that you can bring with you.

With MCP, I think about that as: Underlord is now my collaborator that I bring into Claude Cowork. For example, one of the ways that I create content is I say, "Hey, Claude Cowork, look across everything I've done in Notion and Slack over the last week, and can you come up with 6 ideas for clips that I can make about my thoughts on the AI space?"

This is much better than saying, "Can you just brainstorm 6 thoughts about the AI space that maybe I believe or maybe I don't?" It's like, no, I talk about that all the time.

[laughter]

You've listened. You've been listening. I know you have. Go find some things that I'm saying that you think are interesting, and suggest them for me.

I workshop it in Claude Cowork, and then I'm like, "Great. Go create scripts in Descript. Go create projects for each of these and put the script in there as scratch text so that I can then go into each of these projects and record." Then I can say, "Okay, I've done all the raw recordings. Go turn them into LinkedIn quotes using the skill that I built that tells you what my LinkedIn clips look like."

We have a user I'm obsessed with. Actually, I should send you this Claude skill, but he's been a big user of the MCP, and he has a podcast-editing skill that he's created. He just runs it. It's triggered whenever he finishes a Zoom recording, and it goes through the skill, creates the project in Descript, and he goes and looks at it.

All this is to say that I think we need to bring Underlord not only out of the project and into the drive, but also out of Descript and into the team of agents and into the world where you're already doing a lot of app connections and video. Your video team—Underlord—needs to be in there with the rest of your teams, working on creating your content.

I would love to see the podcast-editing skill. I've created my own. I wouldn't say it's very advanced by any means, but it's kind of a working V1. For example, I typically open every episode by addressing the guest and then saying, "Welcome to The Cognitive Revolution," and usually at the end I've got an outro. I trim everything before the welcome and everything after the thank-you, and there are 5 other steps.

Oh, that's awesome.

It does 6 different things. One is—

Yeah.

It's a pull quote, and then it has the podcast theme come in after the pull quote, and it shows up pretty well.

Yeah, cool. I'm sure I could learn something from that, no doubt, and maybe there'd be somebody who could learn something from mine, although again, it's not that advanced.

I'm sure they could. I think the main thing they could learn is just that you can do this at all, not necessarily the quality with which I've done it so far.

Okay, so this is really interesting, and it gets at some of the biggest questions about the future of software and even the future of knowledge: How is all this going to work?

As an early user of the Descript API—and the Underlord API is really the core of that—I think it's a really interesting pattern that you guys have gone with so far, where the tool is smart. With the Descript API, I am not even afforded the ability to do very specific, fully deterministic edits to a project. Instead, I'm prompting Underlord, and it's doing the thing.

I can prompt it very specifically, but there's always this translation layer. It works pretty well in what I've experienced so far. I'm new to it, like everyone else. I think this is, in one way, maybe the only way for software companies to have any sort of defensibility, because you've got to have some smarts inside your periphery. That seems like it has to be a core principle.

At the same time, there are times when I just want to make sure I'm doing exactly what I want to do. Then there's personalization. My Claude Code universe is ever-growing and has tremendous amounts of context and access to every podcast I've done, as well as ones where I've guested that aren't even in my Descript account. It just has a broader world and a broader view of me.

So how do you see that? I'm thinking about Fin, too. I'm sure you've been following this pretty closely. Intercom has opened up its customer-service model to other companies to build their own competitors to Fin using the same intelligence. Benedict Evans says all software revolutions are either bundling or unbundling. How do you think this is all going to get bundled or unbundled or rebundled? Where are the lines going to get drawn? Where should personalization live? Where is all this going? Please de-confuse me as much as possible.

Yeah, I can't. I think anyone who tells you that they can is lying to you, probably for self-serving reasons. But what I would say is that I think Descript currently—and my job is to make sure this remains true for the foreseeable future—can give you a better experience if you are using Underlord and all of the context that we have about you in Descript.

It's just going to get more and more powerful as we build in personalization and drive-level understanding. I think that is likely to be the primary way that you do video editing within Descript.

If you do it that way, not only do we think that, because of the way the agent coordinates within Descript and understands the capital V capital M Descript model and how everything in the app is set up, we have important context in the Underlord layer, where you're just going to have a better time than if you're asking Claude to coordinate across a bunch of tools.

The other thing that gets you is that if you're doing everything in Descript, then that guarantees that when—which for most people will be true—you inevitably need to go in and do the last 10% yourself before you're really comfortable hitting publish, you'll be able to go into Descript, and we'll have access to all of the discrete things that have been done so that you can undo them and change them.

They're not just flat files that have been put in there. It's like, do you want to edit this thing that's fundamentally uneditable? And you're like, no, this isn't helpful.

I think that's our vision. However, we do think that it is important to break out some of our tools and make them generally accessible. Things like the transcript ought to just be callable in a deterministic way without having to go through Underlord, and there may be additional tools where you don't need a ton of context to be successful with them. We may want to think about distributing those as deterministic tools and giving access to Claude.

But when it comes to really orchestrating multiple types of media, doing visual edits, and setting things up like a layout, I generally think you're going to have a better time. That's true today, and because of all of the work that we're doing with the agent, I think it's going to be true long into the future.

Yeah, I like the framework: “My job is to make sure you have a better time using our thing than using Claude Code.” That's going to be a—

We work together. I actually think you should be using Claude Code, but I think the package that we're talking about is—think about what you're doing—you're telling Claude Code to hire Underlord as your video team, and then to go off and do its job the way that it thinks it ought to be done best.

Yeah. I think that's a good paradigm. And I do think video is one of the areas where—

That is, yeah.

—probably lasts longer than many others.

I think that's right. And look, I am not offended. A common question is just, “How are you going to defend yourself against, I don't know, X, some big lab?” I think the answer that any company that's telling the truth will give you is, sure, if Google or Claude or OpenAI decides that the thing my app does is exactly the thing that they want to be great at, and they want to spend the time and the money and the years and the sustained effort to make a great product to do that job, what can I do? Probably nothing.

But I think that we overestimate the number. I think there will be some low-hanging fruit for them to do that with—a bunch of very lucrative businesses. Having a robust and reliable video editor is actually pretty high up the tree. You have to climb a lot of branches before you're like, “Why don't we just build and maintain something like a robust video editor forever?”

I think that's why a lot of people don't do it.

Yeah, I think the one thing that changes—I agree with that analysis as long as we're in something like the normal regime. Beyond this, it's beyond conventional business strategy. But I don't think it's crazy to think about the possibility that AI coding agents and AI entrepreneur agents can sustain that effort themselves. That's where things get really through the looking glass.

I don't know if you have a view on what, if anything, can be done if that threshold gets crossed. OpenAI, as I'm sure you're aware, has a timeline that they've publicly stated for when they expect to have an autonomous AI researcher, and that is March 2028. We're less than 2 years away from their target to have the autonomous AI researcher.

It starts to be a weird world where you're like, “Geez, that thing might be able to create its own specialist models that could do all these very particular use cases and create its own kind of sense organs to figure out what's going on in the video.” I guess that's just the singularity. I don't know if there's any other interpretation of what happens at that point, but maybe it's just too remote to think about, or maybe you do have some thoughts.

My general thought is that I get really excited when I think about being able to automate more and more labor. I think that generally leads to an exciting future if we're willing to do the work to make it one.

I am skeptical of that timeline and think that when someone tells you something like that, it's really important to get very concrete about what exact bet they're making. Get that on paper, and in the process of getting it on paper, ask: What do you actually mean by an autonomous researcher? What is it able to do without human oversight? What is it not able to do without human oversight?

You often get to a more reasonable picture that still implies a different future than understanding things at the topic-sentence level, which I know you are deeper than that. That's just a general tip that I give people when they're trying to understand claims like, “In 6 months, there will be no more white-collar worker jobs.” Is that what was actually said? Let's slow down and look at the claim that's being made.

Generally, I am very bullish about the direction that labor automation is going in, and I think it's generally good news. We can't predict exactly what the timelines will look like or how it will play out across different industries. That's why I think the companies that will win are going to be companies that are able to make decisions quickly and well, and that generally embrace change and are not resistant to it.

When I think about whether I can tell you right now exactly what the labor market is going to look like in 5 years and where Descript will play within it, no. But do I think that we have built the company rituals that allow us to nimbly shift strategy and quickly take advantage of leaps in labor automation, and quickly build tactics to deal with the competitive situation as it evolves over the next couple of years? That's where I have a lot of confidence.

I tend to think that we are overstating a lot of the near-term changes in labor and society that will come from AI, and probably understating a lot of the longer-term changes in society, culture, and labor that will come from AI. I don't know.

Do you think “podcast editor” is a job in 2 or 3 years? Or maybe people still want to delegate: “I don't want to watch it. You watch it and do a little quality control.” It's not even editing; it's just giving feedback to the AI. That's one version of that I can imagine.

But then it's—I don't know if podcast editor is going to be a job. Will people be employed to tell stories? Yes. What kind of stories, using what media, and who will they be employed by? That's all subject to change.

But if the thing that you're really good at is telling stories for brands, or interviewing other people and finding out what's interesting about that and getting that out—doing a media job—those will still exist. That will still be a job. Especially if you embrace new media and embrace new distribution channels, you may still have that job of the future. Does that make sense?

Yeah, I do wonder. I think there's a huge question around whether people will make these transitions, because a lot of times when I have this conversation, I'm like, “Geez, it seems like this work is going to be pretty highly automated.” Then people point to a different, adjacent kind of work, but a lot of times I'm like, “Yeah, but are the people doing the first thing going to switch to the other thing in any effective way?”

I think that is where I see disruption being pretty meaningful. I feel like the winners and losers are, in many cases, not different people. The fact that there will be winners doesn't necessarily mean that the losers will be able to pivot into a winning position. It might just be a major redistribution of who's winning and losing, and that could be a huge challenge for society, even if there may be all kinds of new things that pop up that do create new kinds of winners and new kinds of opportunities.

To some extent, that's just always true, right? Economies are always changing, sectors are always growing and shrinking, and there's labor displacement. This may happen in a sped-up way, in which case we'll need to make sure that systems are in place to take care of people through those moments of extreme disruption whenever there's tremendous labor displacement in a short period of time, which there may be in this circumstance.

I have a lot of faith that, in the long term, there will be big shifts in labor, but I am not a subscriber to there being permanent losers and irrevocable losers unless that is the society that we choose to build. There may be a very difficult moment where there's an accelerated kind of labor displacement, in which case we need to be ready to meet that moment as a society. But I think it's a moment that we've met before, and changing labor landscapes is not a new human problem.

Yeah, that might tie back to the original slop question in an interesting way. But maybe one more beat on some practical product stuff, because I did want to ask about something that I think is increasingly common and is quite prominent in the Descript experience today. There are individual button clicks, and certainly individual prompts that I can give to Underlord, that will spend a few dollars' worth of credits for me in one go.

And that's a weird new world for software, right? It used to be just click and do whatever you want. Now you're a few clicks in, and you might be through your monthly token budget. How do you guys think about designing for that new cost paradigm?

I think it's a temporary cost paradigm myself. First of all, what I'd say is I personally, as a consumer, hate the concept of feeling like pressing a button is going to cost me a dollar. Even though, actually, if you do a good job creating my clips, that's a pretty damn good deal because that used to take me a lot of time. Sure, I'd love to pay a dollar to get someone to do my clips.

But I'll just say, as a consumer, I don't feel great about that experience. But AI costs Descript money, right? This stuff costs us money, so we can't be free and unlimited. The way that we try to create pricing for Descript is that hobbyists can make one really good thing a month, creators can make one really good thing a week, and businesses can have teams of people making multiple good things a week.

Then it's like, well, what kind of things? Those things really are very different for different people. The reason why it has to be a shared pool of AI credits is that we used to say, “Every creator gets this much AI speech, this much filler-word removal, and this many clip things.” But if I'm a podcaster, I may never need AI speech, and I need clips every single week, multiple times a week. So that's not a good deal for me.

We wanted to have a budget you can spend across any kind of AI job. We looked across all of our main use cases and asked ourselves, “Is this enough credits for someone who's making one podcast a month or one long-form YouTube video a month for a hobbyist?” And for a creator, is it enough for one thing a week? Now you're making how many, did you say, a week? Two a week?

Usually two a week.

All right, you need a double license. But in any case, that's how we tried to price it out. We have the idea that you can add on more credits or more media hours if you're in a special circumstance. That's how we designed pricing, but I think this is a temporary moment in pricing.

Everyone's doing this, and we all know that the consensus around where pricing is heading is toward more outcome pricing, where what you're charged for is maybe something like exports. You're not going to get charged unless you get to the outcome that's getting you the value that you need, and then we'll charge you for that value.

I think we all want to live in that world, but because of the state that the models are in right now and because of how expensive AI is right now, we live in this world that feels uncomfortable for everyone. What Descript does is set up a pricing situation using those general design principles, with the idea that less than 5% of people have to buy some kind of top-up every month—less than 5% of active users.

That feels okay to me. If less than 5% are hitting their limits and needing to buy extra stuff, I'm like, “Okay, this is feeling all right.” If it were something like 50%, I'd be like, “Wow, this is not a fun amusement park to be at. Everything costs so much damn money.”

Yeah, interesting. That's really, I think, a very interesting and useful frame. This has been great. I guess my last question is to tie it back to the beginning and also try to zoom out a little bit: Is it going to be possible to avoid a future of infinite slop?

It seems like right now one barrier to infinite slop is that the models are kind of expensive. So you've got to have some reach, or you've got to have some reason to believe that you're going to get paid back in order to spend all the credits. Maybe in the future we have some sort of universal basic income or new social contract that reduces the need for people to push slop to try to earn whatever pennies per view, or whatever the case may be.

But if pricing is dropping over time, that future may or may not arrive, and that new social contract may or may not arrive in a timely fashion. How do you think the future of content will shake out? Is it going to be infinite generation, and what is it going to look like on the consumption side? Are we going to be lost in slop, or do you have a vision where, even if it's infinite generation on the consumption side, maybe we can somehow rise above that reality? I'd love to hear how you think the future content equilibrium shakes out in an aspirational way.

Yeah, I don't think so. Whenever people ask me about this question, they tend to be people in tech or economists. With a ton of respect for people in tech and people who are economists, I just don't think that what we're missing about content is that content is a little bit businessy. Certainly, a lot of the people who use Descript are using it in businessy ways, but it's a little bit art, too. It's a little bit artistic expression, creative expression, and creative storytelling.

Whenever you're playing in that field, I think it isn't as clearly driven by free-market nihilism as other areas of the world. I'm not talking about Descript now. I'm playfully thinking about video or film as a medium. When something's about artistic expression, art tends to have a way of reacting to the technology of the day and to the culture of the day in ways that surprise us.

I think about the invention of the camera and how that changed painting, the medium of painting, forever. People were just not interested in photorealistic painting after the invention of the camera. I bring that up to say that art always reacts to technological advances in ways that surprise us.

Then you might be like, “Yeah, art—we're talking about content.” But artistic-expression content will change first, and there will be people who do very creative things in this moment that are unexpected, surprise us, and raise the quality bar.

Then there will be businesses that see that and say, “I want a little hunk of that.” It's like that Miranda Priestly scene in The Devil Wears Prada, where she's like, “This designer over here decided that this cerulean blue that was in their spring collection.” Then you buy it in a bargain bin at Marshalls. That's the way that content works, too.

You'll have people who are interesting and exciting, truly creative, and have an aesthetic eye. They're going to do interesting and exciting things both with this technology and in response to and defiance of this technology and this moment in culture. That will inspire all of the marketing people to steal from that aesthetic and that response.

I don't know. It's easy to look into the future and see our nightmares. People pitch me on things like, “There won't even be human creators anymore. What's going to happen is you're going to stare into your phone, and there will be a seed idea of a video. Then, based on where your eyeballs go, it'll generate more and more video in a way that makes you never want to look away.”

I'm like, “Oh, like in Infinite Jest. David Foster Wallace basically told us this would all happen in the '90s.” Maybe that'll happen. It's easy to look into the future and see our nightmares, especially in a world as skeptical of technology as the one that we're in right now.

But I'm actually really excited to see what artists and creative people do with this technology and in response to it. I think that there will be really exciting things that come out of it, and that those are the things that will win in the marketplace, even in a world of slop.

I love an optimistic vision for the future, so I think that's a great note to end on. Laura Burkhauser, CEO of Descript, thank you so much for being part of The Cognitive Revolution.

Thank you, Nathan.