Ivan Burazin

Ivan Burazin is the co-founder and chief executive of Daytona, which builds secure, programmable computing environments for autonomous AI agents. His premise is that agents cannot complete meaningful work without computers they can configure, operate, adapt and reproduce independently.

Burazin began his career installing servers, building networks and working with virtualization. He subsequently co-founded Codeanywhere with Vedran Jukić, developing a browser-based programming environment and the orchestration infrastructure needed to run it remotely.

In 2012, he founded Shift, a developer conference that expanded beyond Croatia. After Infobip acquired Shift, he became the communications-platform company’s chief developer experience officer.

At Daytona, Burazin initially focused on automating development environments for human programmers. In 2024, he open-sourced Daytona’s development-environment manager, which provisioned infrastructure, applied configurations and prepared workspaces. As customers began deploying autonomous software, Daytona shifted toward environments that could multiply quickly, retain state and execute without anyone sitting at a keyboard. The company now supplies composable computers for agents: isolated, API-controlled systems for running generated code, analyzing data and supporting reinforcement learning.

• Agent experience requires operational autonomy. Burazin argues that machine-readable documentation, API-first interfaces and user-mediated authentication should enable agents to finish tasks without repeatedly requiring human intervention. His AI Engineer talk on agent-native infrastructure distinguishes genuine autonomy from developer tools superficially adapted for machines.

• Agent computers should adapt during execution. Conventional virtual machines start with fixed resources; agents may need additional processors, specialized hardware or different operating systems as their tasks change. Burazin has described this distinction publicly, identifying rapid scheduling and orchestration as central engineering challenges.

• Infrastructure should remove practical bottlenecks. Daytona’s declarative image builder lets agents request environments with new dependencies without waiting for someone to construct a container manually. Daytona Volumes share datasets between isolated environments, while snapshots and forks let agents explore multiple approaches simultaneously.

• Execution costs influence agent training. Burazin co-authored research comparing infrastructure for coding-agent reinforcement-learning rollouts, examining how startup delays and execution overhead compound across training workloads.

• Isolation reshaped Daytona’s approach to openness. Although Daytona began with an open-source core, the company closed its production codebase in June 2026, citing security risks associated with exposing sandbox-isolation and orchestration internals. Its public GitHub organization continues to distribute SDKs, integrations and implementation resources.