Is Runway the one to bring an artist’s soul to AI?
Inside the company’s biggest week of the year.
Anyone who knows me knows I’m an enormous cinephile. If you indulge me, I’ll tell you why the 1970s will always remain film’s best decade, and how James Cameron broke my heart with Avatar 2. So you can imagine, as I flew to New York for Runway’s fourth annual AI Festival, I was both excited to be invited and not at all sure how I’d feel about it.
Founded in 2018, when AI could barely generate the crude sphere of an orange falling from a tree (though that blew minds at the time), Runway’s become tech’s best known AI video company. What I saw from the outside was the $5.3B valuation, deals with the likes of Lionsgate, Legendary, Amazon, and Adobe, and the ability to make anything the human mind can conjure. What I didn’t know is that all of this is held up by a relatively tiny team of 180. I didn’t know who they were. And I didn’t know if or how they’d convince me that they aren’t trampling my most beloved art form. All the same, I felt lucky when they invited me to come hang out at the office and get to know them, so I could see for myself.
What I found is complicated. Runway’s technology is truly insane. Some of the video I saw in the office and at the festival is so detailed and spectacular, it stretches the imagination of what can be put on screen. Immediately, I play the doomiest tape in my head: AI drives costs down so much we’re flooded with (mostly) slop and the many thousands of grips, costume designers, actors, editors, cinematographers, etc. etc. lose their jobs.
But that tape doesn’t fit this place or the people I met. After spending time with them and seeing how they’re going about all this, I feel hopeful that AI entertainment is going to sit alongside the cinema we’ve always loved, made by humans for humans - because we crave this too much from the bottom of our souls to let it go away. It’s just that more people will get the chance to tell stories.
Call me naive. Maybe I am. But there was this one moment at the Runway Festival (their annual screening of films showing off new capabilities) where it all clicked. And no, it wasn’t keynoter Ron Howard talking about how technology pushes art forward (though I was as starstruck as any good Arrested Development fan can be). It was Co-founder & Co-CEO Cristóbal Valenzuela at the podium, talking about an AI filmmaker he’d met.
He was an older man who told Cris how, early on, he’d been struck with an idea for a movie. Scraping together the money and resources needed to make it, he came up short. He had to go do something else, live another life. But the idea didn’t leave. It followed him around for the next 20 years wanting to be real. Finally able to realize it with Runway’s tools, he screened scenes at Tribeca and got an amazing response from people in the industry.
I want this for more people. Especially when I saw my favorite short of the night: Costa Verde, a grainy VHS memory of a summer spent with grandparents in the French countryside… and all the magic that resided in those days. I had to embed it for you at the end of this (now very long) intro.
It’s hard to describe the energy in the room at the Festival. Hosted at Lincoln Center’s Alice Tully Hall, it had a scale and polish that fit right in with the Tribeca Film Festival in full swing just 4 miles south. The lobby was humming. The standing-room-only theater roared with excited conversation, hushing to whispers as the lights went down and 10 chosen shorts started to play.
But what made this even better was the behind-the-scenes. The 3 days I got to spend with the team making this possible as everyone flew in for the big show, munching breakfast burritos together, bear hugging colleagues they’d only ever seen on Zoom, changing into their suits and gowns in the office for the evening, just to sneak in a bit more keyboard time in their finery.
The whole experience made it clear that they feel what I felt hearing Cris talk about inviting more storytellers into the tent. In tech, we all talk and think so much about what it means to build with authentic mission and passion these days. Below, I attempt to capture what working at a place that knows feels like.
What stood out to me
Here at Early Days, we aim to give readers a window into cultures we think are fascinating. Having seen so many places on the inside, I want to share what struck me as unusual, singular - the reasons Runway is able to do what they do with so much heart.
When you walk into their NYC HQ, you’re immersed in creation. Every screen in every room is glowing with videos made by in-house artists. You see pigeons arguing over a time machine, pizza slices doing the conga, a smiling martini, a little robot with a letter he must deliver. “Everyone is making things for the Slack #generation channel all the time,” one of their technical artists tells me. “All the gems go into sizzle reels.” Almost all of them are trying out new techniques to see if they’ll work. Almost all of them are inside jokes that you can cite to give anyone a case of the giggles. There are so many backstories in this place.
They take crazy, massive swings (no really). Even in 2021, back when the best they could do was a 500 x 500 pixelated vid, the founders knew they’d be showing real films on the big screen, first engineer Yining Shi told me. Knew it with such conviction they plotted their first film festival for the next year. Now, they’re charting equally ambitious courses into world model development, and generative UI, building models that can simulate increasingly large parts of the real and digital world, with the hope of solving some of the most pressing problems for humans. More on these later (both extreme bids), but they’re taking them on with the same conviction that now has them filling a massive theater just 5 years later.
Quick break to tell you a related story: When OpenAI released Sora in 2024, the whole world seemed to say: Runway is done. “It was a full-blown existential crisis that lasted… 20 minutes,” CTO Kamil Sindi says. “Because we didn’t have time to panic.” Co-founder and now Co-CEO Anastasis Germanidis told him the team needed to not just match but leapfrog Sora’s performance in 60 days. “It was honestly crazy. We were way behind and didn’t have the infra we needed.” But they did it, blowing past benchmarks. They did it with daily standups, no ticket writing, and a ton of trust. The result was Runway Gen-3, and looking back Anastasis now says: “It was actually the best period in our history. Morale was huge after that.” It’s a story several people tell me. It clearly lifted the lid off of what could be done.
Started as a thesis project, founded by 3 artists. Cris’ university had the largest private film collection in Chile with a rental limit of 2 DVDs a day. He maxed them out. Alejandro ran his own design studio in Chile at age 19. Anastasis grew up making music, and went into interactive theater. The three met on the same floor at NYU’s Interactive Telecommunications Program, an experience that clearly broke everything open for them. Cris started Runway as his thesis. The other two soon joined. “Anastasis was a scientist who wanted to make art, Alejandro was a designer who wanted to be an engineer, I didn’t know what I wanted to do.” So he built this, setting the DNA for the company to come - artists, researchers, and engineers all together in one room making something together.
They make remote work better than anywhere I’ve seen. Remote is now fully out of style, but here, it’s worked (internationally, too!) from the beginning. Runway’s third hire Bryan was in New Zealand. They never met. Cris had just seen his work online. For them it’s all about finding the most exceptional people, wherever they might be. There’s one thing that makes this possible from my perspective: They have the richest, silliest, most engaged Slack instance ever. I got snuck an invite by Cole Garry in comms, lurked for a week before my visit, and I couldn’t believe it. Anyone can post anything in #general - even if it’s fancy coffee in the kitchen. Every post is met with a deluge of emojis, including many custom emojis. Every employee gets a custom emoji of their own - PM Ryan has his face x5 because he’s always doing 5 people’s jobs at once. Kelly is coronated with a crown. Profile pictures change at the speed of memes - the goal always being to make someone laugh as hard as possible. There are robust and eclectic #random, #overheard, and #ideas channels. There’s a casualness, a lack of polish that to me reads as comfort. I’m told by a 23-year-old engineer that anyone can ping the founders and they’ll respond right away. Another tells me “I never get an adrenaline spike when I see a channel light up here.” All of this seems like it should be normal… but it really isn’t.
There’s a wealth of lore. This is a squishy one. If you asked most founders how important it is that people tell stories from “back in the day,” many would say not at all. But Runway is thick with nostalgia for that first Bushwick warehouse, the tiny cluster of desks where they hashed out the basics, their daily after-lunch Sey Coffee walks where they talked about the big things they could do. The next generation speaks wistfully of their last office, the one in Tribeca where “we all grew up together” an exec says, and how after they shipped their first video editor, Sequel, the founders sent everyone plants with a card: “We just planted a seed.” When asked to see photos from those days, I’m presented with a Polaroid
teamfamily album. Brie and I talk about this one a lot, companies that have lore and those that don’t. It’s more and more obvious. It’s an invisible force that holds people together.
Some odd ways they build their team (almost like it’s art, too)…
Six people I spoke to said they joined because Cris saw something they made on the internet and DM’d them. He’s always trawling GitHub, HuggingFace, and YouTube for new stars.
They don’t sell hard. Even when a candidate is one of the AI elite, if they pit Runway against other offers or aren’t as excited, the recruiting team doesn’t chase. “We want people who want to be here. The biggest problem with the big labs is they tend to hire all crazy superstars,” one engineer tells me. “But that doesn’t mean they work well together.”
They started caring about the humans unusually early. Part of this is because of Michelle Kwon, the first business hire who is now COO nearly 6 years later. Even when they were in that Bushwick warehouse, she was already figuring out parental leave and 100% 401k matching. They hired a high-flying Head of People, Anna Chalon from Adobe at just 25 people, who furthered these practices, which Michelle calls being “smart with heart.” These two women have so much obvious adoration for each other, when I asked why they love working at Runway, they point at each other. (This care now extends to Cris pinging Anna things like: “I heard X person broke their leg, did we send something?”)
Onboarding is diving right in. For a long time, the smaller team took every new hire out to dinner. One person’s first day was the same as their Series C close, “I think everyone went without me and I stayed behind with the team doing the deal,” she says. “We got sushi, and soy sauce was everywhere.” This has morphed a bit - you’re still thrown into the deep end, but every new batch of hires now gets an unvarnished AAC meeting with Anastasis, Alejandro, and Cristóbal. I went to one, and it really is new people hurling extremely hard questions at these guys: If growth stagnates in a year, what went wrong? What 10x bets has Runway chosen not to take? Several times I wince, but the founders field everything without flinching, answering in a way that already extends huge trust.
This one’s my favorite: Runway tenures are long. 8 of the first 10 employees are still here. And people step up and up and up. Michelle is just one example. She speaks with a purpose and precision of that office badass who knows where all the bodies are buried, but also cares if you’re warm enough. Some places, a career COO might have been brought in. At Runway, she’s always been running the show, so she is now. It speaks to a loyalty, respect, and belief in people that’s too rare these days. The other one I absolutely love is Kelly Chase, who before joining worked for 10 seasons at a seafood shack in Cape Cod. COVID set in and she decided to get a customer service job - applying for 50 different roles. Runway was the one she got. Cut to today. She runs the Festival. She produced their book of values (Meditations on Making the Impossible), interviews artists for their magazine, Telescope, and jumps in on all sorts of creative production. She’s the beating heart of the place. “I used to drag my friend every year to the New York Film Festival,” she says. “This year I’m organizing our festival in the same place.” The room was full and you could see her fingerprints everywhere.
A quick aside I don’t know how to fit anywhere else, but must include: Several people I speak to say they look better and feel healthier now that they work here. One even shows me her Oura ring data. “All my friends tell me my skin is glowing now…” says another, “though I wonder if I looked like crap before.” I’m dubious… but it’s spooky how consistent it is.
This last one may sound kind of cliché, but not how it shows up at Runway. They weave customer love into as many aspects of employee experience as they can. Slack channels, brainstorms, and my favorite of all: the “Runway Love” segment of All Hands. It stars PM Joel Kwartler, off-hours standup comic, who delivers comments from customers a different way each time (interpretive dance, a theatrical recitation). The one I saw, he and some fellow be-wigged bandmates held a sudden rock concert - with quotes as the lyrics. The whole company, in town for the Festival, cheered along. Brie says it’s the best All Hands she’s ever been to.
Increasingly, Brie and I have a working theory that the most powerful company talent brand stems from employees who feel so strongly about internal culture and the impact of their work, they share out loud in the world. There’s nothing more motivating than marinating in effusive customer experiences. Sure, they get criticism too, but leadership knows the love is fuel.
Speaking two languages
User love is easier when your ideal customer profile is sitting side by side with your team. In this case, digital and visual artists are seated not just in the same vicinity, but next to the researchers training the models. “Every day you see researchers talking to VFX artists with 20 years of experience,” CTO Kamil says.
As you might imagine, he explains, taste in video models is very hard to measure. Deeply subjective. Why does a certain video just feel better? How do cameras move naturally? It’s more complex when you think about your favorite movies. Multiple angles of the same subject blend so easily, there’s continuity of style. The artists on staff are much more capable of realizing, in a split-second frame, that strands of hair are not moving the right way in the wind.
Two product managers gush over an internal tool they pioneered for artists to answer for every frame: is this good? Why is it good? Why is it not good? They get into the one-by-one breakdowns of it all, and often sit down for long, in-person critiques.
But they’re much more instrumental to the end-to-end than running evals. They’re also more aware of how weird, unexpected, high-signal data should be collected to get to the right results. To create their advanced “Multi-shot” feature inside their editing model Aleph 2.0, visual artists encouraged a group to play soccer on the roof with GoPro cameras on their heads. The model needed to see how to capture ball movement from many perspectives. They put cameras inside the fridges looking out, around the kitchen as people made cocktails. They knew how to get to data that’d capture real-looking movements, surfaces, and gestures. This data set didn’t exist anywhere else in the world before.
“At most companies, there’s this waterfall where research builds, product packages, and creatives get to try it out - we don’t have time for that,” says early engineer Yining.
She’s working on a product called Characters, which needs to generate video avatars capable of conversing with humans in real time. She enlisted artists to help her understand cadence, facial movements, how quickly characters needed to respond. It’s gotten so good, they were able to mock a fake version of the host of a recent conference they presented at, baffling him as the whole crowd roared.
“The people who thrive here speak two languages: art and science,” says Anastasis. “But they have to speak it at the same time. Our first engineer ever went to art school and now he’s this epic systems backend eng.”
Their place in the process is reflected in how Runway shows up to customers. “Sales is never like, ‘yo thanks, we’ll take it from here,’” another PM tells me. “Honestly, sales usually sits in the back of the room and lets the artists present to people just like themselves. We try to put them at the center. The things they see and know, you can’t fake.”
Ian Sansavera is the platonic ideal of a Runway visual artist. He went from making prank videos as a teenager in Ohio to producing vlogs for UFC fighters, to teaching 300,000 YouTube subscribers on his aptly named channel “Learn How to Edit Stuff.” As per usual, Cris spotted his Adobe Roto Brush tutorial and reached out. Now at Runway for 5 years, he’s the ringer you bring to meetings to show off the tools in their native environment.
At meetings, he’s regularly recognized by people like the head of VFX for the original Avatar movie. That’s the ilk that followed his YouTube channel. He found himself in a sales pitch with an Oscar-winning effects creator working on something new. He got the guy to AirDrop all the files so Ian could show him around the early version of a new model with the real material. He also commandeered Runway Academy, tutorials the company puts out, data curation, and started the famed Runway Discord, now home to 20,000 hyper-engaged members.
Curious how this worked on the customer side of the table, I caught up with filmmaker Dave Clark, who entered his dysmorphic influencer short “Tairell isn’t Real” to the festival. He’s been creating with AI for the last 4 years, and even founded Promise, his own firm making video for others. He says it’s clear how much nuanced care has gone into making the product dead simple for creators.
“I can talk to the model the way I’d talk to my crew on a film shoot,” he says. “None of the newfangled terminology, just ‘the camera sprints toward the captain,’ ‘it’s shaky like we’re crawling through the mud.’ I love that that’s how the AI wants to be spoken to.” It’s not an accident this is the case. “We’re crossing a threshold of people caring more about the story and less about how to get things to work.”
Dave’s excited about how Runway helps filmmakers prototype and derisk ideas before going out to shoot more traditionally. “I’m really hoping it reduces studio nervousness around risk-taking on IP and new filmmakers. That’s the win I want for the whole industry.”
Dave’s an example himself of how this is working. When he shopped around a script requiring elaborate battle scenes dipping above and below a chaotic waterline, he got comments like: “love the story but can’t visualize it.” He ended up making the movie with Runway, only to get a call from Ridley Scott’s office asking: “How did you do that?” His inner child who’d watched Alien and Blade Runner next to his dad rejoiced.
There’s something about the concentration of artists and art lovers in this space that magnetizes the right people. I regularly hear about unhinged compensation for the world’s best researchers - a dynamic that’s made it hard for application-layer companies looking to compete. In this reality, it can feel a bit cheugy to say, “mission matters!” It doesn’t feel solid enough to pull talent. But I see it working at Runway. They’ve created an environment so specific that they can outcompete for candidates who value art and want to be around their people.
Building new worlds
These days, researchers are becoming more important than ever to Runway’s trajectory. Starting from their foundation of video, they’ve been steadily growing into world models, with an eye toward robotics and autonomous tech. They have a somewhat controversial theory on this, as relayed by Daniel, the youngest engineer at the company. He refused to drop out of college because, as he put it, “you only get to be a kid for so long.” But at Runway, he’s working on a very adult-sized bet: video models should become the brains for navigating the real world if you have enough footage - we’re talking hundreds of millions of hours - then you should be able to teach something to act the right way.
To win, Anastasis says, they need to build the world’s most powerful base video model. How does this differ from what else is out there?
Most companies are trying to teach robots by putting them in the world: in kitchens, warehouses, bedrooms, labs, anywhere they can collect more demonstrations. Runway thinks the internet is already full of evidence about the physics of the world: hands opening drawers, cars turning through rain, people walking dogs, glass falling off tables, kitchens being cooked in, cities being crossed. Enough video, in their view, becomes a kind of sensory inheritance.
Daniel talks about model training like baking a 100-foot-by-100-foot cookie in your kitchen. You only get so many chances at the real thing, so you test the recipe small many times first: “Can the model learn motion earlier, can it learn physics earlier, what does it actually absorb about the world?”
If a model can learn motion, continuity, causality, camera movement, human behavior, and object permanence, then “AI video” starts to look like the small end of a much bigger wedge. From there, the same foundation can extend into gaming, real-time characters, generative UI, robotics, autonomous systems - anywhere a model needs to understand what is happening in the world and what might happen next.
Early signs are encouraging. They’re seeing strong results in real-to-sim correlation of their robotics world model, and they’re lengthening context windows, but they still need to prove that this is going to have really valuable real-world applications. Kamil walks me over to the office’s newest addition: a bulky white robot with pincers for hands named Abdul, which they’ll use to start testing their theories. Two more are on the way.
The real challenges are around compute demands, speed, and costs. In short order, they need to build initiatives around diversifying their data sets while compressing compute and latency. This is where the harder core technical side of the business becomes apparent. Training frontier-scale video models is one of the harder (and most expensive!) problems in tech. In our estimation, there’s only about 5 contenders in this game.
First, there’s the bet they’re placing that they must bring in multiple data streams at once: video editing, real-time avatars (Yining’s Characters project), gaming and world creation, and the physical intelligence of robotics. This is a lot for a small team to do. They’re hiring against it fast, but everyone here feels the stretch to get to a more capable, polymathic general model.
At the same time, they’re working on generating video autoregressively, frame by frame, fast enough to run live and stay coherent for longer. Every second requires 24 frames at a minimum - gaming, sports, and other complex motion requires more. And that’s before audio processing (a deep research problem in itself) and network lag, so any slip kills the sense of “liveness.” You end up in a very uncomfortable uncanny valley. The standout breakthrough here has been GWM-1, their world model that builds an internal sense of physics and human expression (the underpinning of Characters). But this was clearly just a milestone in a constant, unrelenting drive toward more audio-visual realism, and the better data preparation, enrichment, and fine-tuned debugging it requires. (One of the company’s most valued engineers, Wei, is known as a preternatural debugger who has an instinct for it that, by all accounts, outcompetes Claude.)
There are also some big infra boulders to move. For years, Runway has been working on its own inference engine to make video generation cheaper and faster, Kamil tells me. With such big models, the team has to chop them up and spread the pieces across large, distributed GPU clusters. They’ve had to create internal tools for granular data tagging. A tool to route compute to teams so no GPUs stand idle. Workers that pull weights straight from peer GPUs that are already loaded to turbocharge cold starts.
There’s a lot that lives downstream if Runway can get the base video model right. I sit in on a research brainstorm with 23 people dialed in from everywhere asking what happens in a world where all UI is generated video. They show an interactive map of New York attractions, a frog typing responses, happy RNA molecules showing how proteins are made. The software should compose itself around your intent. Websites and apps will become whole worlds that respond to users. Getting back to artist roots, some of these experiences might be stories that people can inhabit, explore, and act within as an entirely new type of media.
Suffice it to say, this type of breathless sci-fi postulating is very normalized at Runway. At many companies I’ve seen, new ambitious initiatives have felt bolted on. They feel like bloodless imperatives just to make more money. But there’s a through-line here. Anastasis believes that a great video model is the best path to a great world model (which, he tells me, is the next big breakthrough AI has in store). Getting there first means more than just grabbing a slice of markets like robotics and autonomous vehicles. The Runway team is full of true believers that these markets will be fundamentally different because of their work.
All of this has necessitated a beefing up of the go-to-market side of the house. As they talk to more enterprise customers about new offerings, these new muscles are getting a workout.
What’s interesting to me about this is that I don’t see the typical strife brewing between B2C and B2B sides of the house. It’s amazing how predictable this has become. As soon as enterprise really starts printing money, the sun shifts away from the IC users, the community that arguably got the company to where it is. But Runway already has a significant B2B business and is showing no signs of abandoning their Creative Partner Program, their Discord obsessives, the people out there making things every day, so excited to show what they’ve made. While I don’t meet him, I hear many reverent tales of Timmy, Discord moderator, who keeps the masses happy and seen.
It might be too early for such a schism to take shape here - 180 people is still small, after all - but I hope, given all the other unique aspects of this team, they find an even better path to walk this line than their PLG predecessors.
Season tickets
For a week after my visit, I was so enamored with the people I got to eat breakfast with, sit next to, watch screen their stories, I started sending Runway candidates and telling all the talented people in my life they should really take a look at their careers page. But my brain was still buzzing with something else: a hope (however cautious) that they’ll be the stewards of this technology that they appear to be today.
I’m that friend hoovering up every scrap of info on AI and Hollywood. The loathing, the fear, the “Fuck AI”s from Guillermo del Toro and Jacob Elordi. The tentative intrigue coming from Ron Howard, Martin Scorsese, Darren Aronofsky. A24 saying yes to an investment from DeepMind. The push pull of: “look at everything we can do now!” vs. “it’s inevitable - deal with it” vs. “we have to stop it!” Honestly, I don’t have it fully reconciled, so for now I’m watching.
I will say, however, that in this watching, I have Runway season tickets. When Brie and I set out to write these profiles, we knew we’d be positive. At the same time, we decided to only write about companies we believe in and let that be the editorial statement. Runway fit this bill for me as soon as I sat down with Michelle and Anna on my first day and heard them talk about the team they’d helped build. I only got more sure as I spent time with the founders, and the people that feel they all grew up together these last 8 years.
What I saw from all of them - the genuine desire to make the world a more vibrant, inviting place for stories - makes them the people I want at the helm of this seismic shift, and everything that comes after that.
Join me in the front row?
Hi Cam, Happy to connect! I sent you a DM here on Substack.