Composable computers for agents: A conversation with Daytona CEO Ivan Burazin
Daytona CEO Ivan Burazin explains why every AI agent needs its own computer, how he built an AI with its own identity, and why SaaS is shifting from seats to consumption.
Daytona builds what the market calls sandboxes. CEO Ivan Burazin has a different framing: composable computers for agents. At HumanX 2026, he sat down with WorkOS CEO Michael Grinich to explain why that distinction matters — and where it's all heading.
Agents need computers, not just APIs
Humans use different computers for different jobs. A MacBook Air for light work, a Linux box for development, a Windows machine for 3D rendering. Burazin's argument is that agents need the same thing: access to different computing environments matched to the task at hand.
"Every single agent in the world will need a sandbox for every single task it's doing," Burazin said. "As agents are able to do more things like humans, they're going to need a similar set of tools — and those tools are usually on computers."
The product thesis Daytona is building on is that the more capable agents become, the more they need full computing environments — not just API endpoints, but actual machines they can operate.
The pivot that caught the wave
Daytona pivoted to sandboxes on January 15th of last year. But the go-to-market started before the product shipped. They invested heavily in in-person events and brand awareness — including those San Francisco billboards that were hard to miss and a major conference at the Chase Center.
The strategy was deliberate: build name recognition first, then launch the product into a market that already knows who you are. By the time Daytona's sandbox product went live, the brand had momentum even if most people didn't yet know what Daytona actually did.
The human emulator: an AI with its own identity
The most compelling part of the conversation was Burazin's personal experiment with what he calls the "human emulator." He set up a Claude-based AI agent with its own sandbox, email address, phone number, and accounts across Daytona's internal tools. The agent operates autonomously inside real systems.
When the agent encounters missing data in a report, it doesn't surface partial results and wait for a human to fill in the gaps. It logs into the relevant web application and retrieves the data itself.
"It is literally equal to a remote person you're working with," Burazin said. "Because it has access to all the same tools — and it needs a computer to be able to do these things."
The setup is nontrivial, though. Creating what amounts to a "fake person" — with a unique email, phone number, and distinct identity across multiple systems — takes effort. And the agent has an unexpected failure mode: it sometimes confuses itself with its creator.
"Sometimes I have to fight it and say, 'You're actually your own person. You're not me.'"
It's a strange problem to have. But it reflects something real about where agent capabilities are headed. When an AI operating under its own credentials begins defaulting to its creator's identity in conversation, the boundary between "agent as coworker" and actual coworker becomes an engineering problem, not just a metaphor.
From people-majority to machine-majority
Burazin's vision extends beyond individual productivity. He sees a structural shift in how software companies should operate: every SaaS company should become an API company, charging for consumption rather than seats. The reason is straightforward — the ratio of human to machine interactions is inverting.
"We're going to go from people-majority to agent-majority," he predicted. "It's already happening. It's going to be like 90%."
If that ratio holds, per-seat pricing breaks down. When most of your "users" are machines running tasks autonomously, the unit of value isn't a human sitting in a chair — it's compute consumed, actions taken, tasks completed. Companies that don't adapt their pricing models to this shift will leave revenue on the table, or find their products used heavily by agents while only a handful of human seats generate revenue.
That's the bet Daytona is making: that the infrastructure layer for agent computing is about to become as fundamental as cloud computing became for web applications. Every agent needs a computer. Daytona wants to be the one providing it.
This interview was recorded at HumanX 2026 in San Francisco.