Moving fast and tackling complexity: building systems that scale at OpenAI
OpenAI is at the forefront of artificial intelligence research, developing systems that can understand and generate human-like text, code, and images. Founded in 2015, they’ve grown from a small research lab to a technology pioneer with over 2,000 employees, shipping products used by hundreds of millions worldwide.
When you’re small, friction is manageable. A two-person engineering team can keep track of their work in their heads, coordinate through chat, and still ship quickly. But as organizations scale, these small points of friction don’t just add up, they multiply.
“Ship yesterday” characterizes the engineering organization at OpenAI. Engineers might work on eight different products in a single year as new challenges and opportunities emerge. At this speed, even small friction can feel like you’re sprinting with a wind sail on your back.
When you’re building products that will be used by hundreds of millions, these points of friction can have a compounding effect.
Tackling friction effectively requires both the right systems, and an appropriate culture around them. From a small trial, to over 2,000 people across the organization, teams at OpenAI organically adopted Linear to help them navigate the complexities of scale–from keeping teams aligned across complex dependencies, to maintaining speed as the organization grew. We spoke with engineers at OpenAI to understand their first-hand experience of using Linear.
Complexity as the status quo
“It’s like an archipelago,” explains Gabriel Peal, an engineer at OpenAI, describing his regular experience working at companies that didn’t use Linear. “Every team is on their own island, using their own tools and systems.” Projects lived in multiple tools, and many existed only in engineer’s heads.
Teams chose tools that work for their immediate needs, optimizing for local efficiency over organizational coherence. This result is a maze of disconnected systems that makes collaboration increasingly difficult.
“If you went and assigned an issue to another team,” Peal recalls, “you didn’t really know how to do that, and things would get lost.” Each handoff between teams became a potential point of failure. Every cross-team collaboration required navigating different workflows, different labels, different ways of thinking about work.
This type of challenge is particularly acute for teams building critical infrastructure alongside external partners–where coordination between teams is most important, and the impact of stale statuses or things getting lost is most heavy. Atty Eleti, who worked on OpenAI’s Apple integration explains: “Once you cut a version of an API, you have to wait until the next major version to make changes. With Apple APIs in particular, these are on device, and once a device API is shipped, you have to support it for several years. It’s really important to nail the details.”
In this environment, even small oversights could have long-lasting consequences. Teams weren’t just managing their own work–they were trying to manage the spaces between teams, the handoffs, the dependencies. It was a tax on every interaction, a friction that grew with each new connection.
Simplicity scales
The change at OpenAI started small, with individual teams choosing to try Linear. “It’s like a Katamari Ball,” Peal describes. “You get a couple of people to use it and then it snowballs.” From an initial hundred seats, usage grew to over 2,000 people across the organization–and as Peal notes, “it still feels perfectly performant. Search hasn’t slowed, it hasn’t become harder to find things, it’s still fast and simple.”
What drove this adoption wasn’t an abundance of features, quite the opposite. While other tools pride themselves on flexibility, offering endless customization of fields, workflows, and processes, this “freedom” often becomes a burden.
Often tools have a lot of features. This often means people configure a lot of customer workflows, which end up making the tool super slow and hard to use. People shoot themselves in the foot because they do too much.
The paradox is that too much choice often leads to fragmentation. When every team can create their own unique ways of working, they do–recreating the same islands of disconnected process, just within a single tool.
Linear takes a different approach. By focusing exclusively on the craft of building great products, it’s opinionated about how work should flow, offering a thoughtfully curated set of features rather than endless options. This isn’t about restriction–it’s about creating a framework that helps teams think in scalable systems from the start.
“If you think about it ‘the Linear way’, you’ll figure out how to do it in a way that works well,” notes Peal. The result isn’t limitation, but liberation. Teams spend less time debating workflows and more time building. They focus less on managing their tools and more on using them.
This approach recognizes a fundamental truth about scaling an organization: simplicity scales. The best systems aren’t the ones that can do everything, but the ones that help teams to do the right things naturally. It takes careful design to create something simple that works.
Using tools that are such high quality sets a standard for you to carry into your own day-to-day work. It’s almost like setting a watermark. Subconsciously affecting what it means for something to feel fast, or have thoughtful animations that actually add to the experience. You’ll probably build a better product, just because of the craft that using Linear infuses on your brain.
When culture meets the right tools
Culture builds slowly, through small actions repeatedly daily. At OpenAI, where speed defines everything, this presented a particular challenge: how do you maintain good habits when moving incredibly fast?
“Every standup, we would pull up Linear, we’d go through the project, we’d screen cast it, and people would talk about tickets that were ongoing.” Eleti shares. These weren’t dramatic changes–just small, consistent cultural practices that developed over time. Linear is just the enabler.
Tools alone don’t make a system, it requires the right culture around them. However, the tools we use matter more than we often acknowledge. When updating tickets feels like a chore, when finding information requires diving through complex customizations, when every team uses different workflows–you’re trying to build good cultural habits on hard mode because your tools are working against you.
It really matters that the friction to create a ticket or the friction to modify a ticket, is really low. Otherwise, people won’t use it.
The reality is subtle but important. While good tools alone don’t create good culture, bad tools can silently kill it. Each moment of friction, each small frustration, builds invisible resistance to practices that could otherwise become natural habits.
At OpenAI, the combination of intentional practices and low-friction tools created something powerful. Teams began naturally staying on top of their work. Not because they had to, but because it made their work easier. Good habits formed not through mandate, but through the gradual recognition that clarity and coordination made everyone more effective.
This is the kind of change that’s hard to measure but impossible to miss. It shows up in the small moments. Engineers automatically updating tickets after conversations, teams having clear visibility into dependencies, projects moving forward without confusion about status or ownership. None of these alone transform an organization, but together they enable teams to move fast without losing their way.
It's a feeling
The best tools aren’t the ones that do the most things, but the ones that help people work together most naturally. As OpenAI continues to push the boundaries of AI, they’re finding that the key to moving fast isn’t just about having more horsepower, it’s about building systems that reduce friction in how people collaborate, one detail at a time.
What makes the right tool isn’t always easy to explain. When asked why other teams should consider making a similar switch, Peal’s response: “You just have to use it and you’ll see. You’ll just feel it.”