Rita Kozlov from Cloudflare: competing with the hyperscalers
Cloudflare knew it could not beat AWS at operating the same centralized compute model. Rita Kozlov explains how Workers made a different, orders-of-magnitude-more-efficient game possible — and why the company uses its own platform to keep shipping like a startup.
Rita Kozlov is VP of Developers and AI at Cloudflare. She joins Jack just after the launch of Cloudflare Containers to explain how that missing primitive fits into a platform originally built around Workers.
The conversation ranges from disruptive technology and long-term trust to product feedback, infrastructure release discipline, small acquisitions, agent workloads and the cultural mechanisms that keep a 5,000-person company moving quickly.
The Q&A below is experimental — extracted from the episode transcript with AI. If you spot a mistake, let us know at jack@scalingdevtools.com.
Do containers complete the Cloudflare developer platform?
In a way, yes, because customers and our own teams have a long tail of packages that need somewhere to run. The container system had already powered products such as Workers Builds and Browser Isolation internally. We debated exposing it directly for years because we did not want its familiarity to obscure what Workers already made possible, but after proving a platform on isolates, containers could fill the remaining gap.
Why did Workers give Cloudflare a credible way to compete with hyperscalers?
We originally needed customers to program a CDN, where ten-second cold starts would make no sense. Isolates produced compute that was dramatically faster and more efficient. We could not beat AWS or GCP by copying the economics of enormous centralized data centers, but a new execution model changed the game. It also delivered more fully on the serverless promise by removing the need to manage regions, concurrency, cold starts or machines built only to keep functions warm.
How does disruptive strategy show up in everyday product decisions?
Cloudflare has leaders who studied Clayton Christensen directly, so the question of what unique long-term advantage we bring is part of the company DNA. We repeatedly take capabilities once affordable only to enterprises and make them broadly available. A new primitive should not merely fill a checklist; we ask how it lets us compete over the long run and whether it makes previously costly technology accessible.
Why does Cloudflare make so much infrastructure free?
Trust is the foundation of the business. Customers let us decrypt and inspect traffic to protect them, so short-term actions that weaken trust would make the long-term company unsustainable. That thinking applies to transparent incident responses and to a free tier that is not a temporary trap. Making something such as SSL broadly free may cannibalize a direct upgrade, but improving the Internet can still be better for Cloudflare over time.
Why are so many Cloudflare product builders visibly online?
We hire people who care deeply about their work, and those people naturally want to know what users are saying. PMs and engineers monitor Twitter, Discord and other channels, respond, and learn from them. With AI changing daily, staying close to the zeitgeist also helps the team understand which models and products matter. Many excellent builders are not public, but visible curiosity and care are not accidental.
How does Cloudflare keep shipping quickly as a large company?
Our first advantage is that we built a developer platform and use it ourselves. The second is a culture of prototypes, MVPs and weekly demos. Shipping creates a stream of feedback that both improves the product and motivates the next release; withholding something for too long can increase the chance of solving the wrong problem. Teams should not get trapped indefinitely in ideation or an impossible search for perfection.
How do you reconcile speed with infrastructure reliability?
The release process depends on maturity. A new product with zero users gains nothing from sitting in a 48-hour canary. The Workers runtime can affect live websites, so it uses feature flags, opt-in compatibility behavior, canaries and gradual rollout. Even beta labels do not make breaking changes harmless because somebody will rely on the first API. Judgment and instrumentation matter more than applying one dogmatic process everywhere.
Why is Cloudflare well suited to running agents?
Agents combine real-time interfaces, long-running model calls, memory and human-in-the-loop pauses. If users pay for every idle moment, running hundreds or millions of them becomes infeasible. Cloudflare's real-time products, Durable Objects, flexible execution and pricing can support those distinct patterns. The Agents SDK packages memory, UI and MCP helpers into a lightweight library so developers can start without assembling every primitive themselves.
How do small acquisitions help Cloudflare?
People are the lifeblood of the company, and startup teams bring a strong bias toward finishing things. We look for curious, humble and empathetic people with a vision strong enough to move through obstacles. Acquiring companies already built on Cloudflare also avoids the common integration failure where incompatible technology consumes the deal. Sometimes a relationship starts with a founder publicly finding bug after bug until the obvious response is: do you want to come fix them?
What does Cloudflare do that more startups might copy?
Innovation weeks create a deadline for work stuck at 90%, but they are grounded in the mission rather than launch theatre. Birthday Week has shipped things such as free SSL for everyone, turning an enterprise capability into a public good even when the short-term revenue logic was not obvious. The weeks can be chaotic and come together late, but they force ambitious ideas across the line and keep the mission visible.
Links & references
- Rita Kozlov's X ↗
- Rita Kozlov's LinkedIn ↗
- Cloudflare ↗
- swyx's article on Cloudflare ↗
- Stratechery on Cloudflare's disruption ↗
This episode is brought to you by WorkOS.