Daytona says people managers make better AI-agent operators

Why it matters

Burazin's model turns management skill into a technical advantage: small teams can run more parallel work, but only when humans specify outcomes and judge machine output well.

Ivan Burazin (@ivanburazin) says Daytona's strongest AI-agent operators share a useful qualification: they have managed humans before.

The Daytona co-founder and CEO told Business Insider that employees with management experience tend to give agents clearer inputs, define outputs more precisely and communicate how they expect work to be completed. Daytona has about 30 employees across the US and Europe, Burazin said. Its 16 engineers each run an average of five agents, and, according to Burazin, nobody at Daytona writes code directly anymore.

That account is a founder's observation from inside one workplace, rather than a controlled comparison of managers and individual contributors. It still offers a concrete picture of how one AI infrastructure founder is reorganizing software work: engineers are becoming supervisors of parallel machine labor, and their output increasingly depends on how well they specify, review and correct that labor.

Burazin has spent much of his career building tools for developers. He and Daytona co-founder Vedran Jukic started the browser-based development environment Codeanywhere in 2011. Burazin founded the Shift developer conference the following year, then became chief developer experience officer at Infobip after it acquired Shift in 2020. He started Daytona in 2023 with Jukic and Goran Draganic, returning to the development-environment problem with a different primary user: the AI agent.

Management becomes specification

Burazin's observation rests on a basic management habit. Experienced managers know that assigning work requires defining the desired result, supplying context and setting boundaries. Individual contributors who are accustomed to holding the full problem in their heads can assume that other people, or agents, will infer what they mean.

"People who manage people and understand how to communicate what they want" are better at running agents, Burazin told Business Insider. He said less experienced managers can become frustrated when agents fail to produce an outcome that was never fully specified.

Burazin is not proposing a return of managers who only coordinate other people's work. He told Business Insider that managers without practical knowledge in their function are not useful. Under Daytona's model, a technical manager still needs to understand the code, product or campaign well enough to judge an agent's work, identify a weak result and redirect the process.

That distinction matters because agent supervision can disguise poor work behind high activity. Five agents can produce five times the drafts, branches or pull requests without improving what reaches production. Daytona's reported figures measure agent use. They do not measure code shipped, defects introduced, review time or customer outcomes.

Research does support the narrower idea that how people instruct AI systems affects results. A May 8th Columbia Business School research summary described experiments involving about 3,750 participants and roughly 37,000 prompts. On a task with a precise target, adapting prompts accounted for nearly half of the measured improvement associated with a newer image model. The effect was much smaller for open-ended creative work. The research did not test whether people managers outperform individual contributors.

Daytona is testing its own product thesis

Daytona gives agents isolated computers where they can execute code and use development tools without exposing a customer's underlying infrastructure. Its documentation describes sandboxes with dedicated kernels, filesystems, network stacks and allocated compute. Agents can create environments through an API or software development kit, run processes, preserve state and resume work from snapshots.

The product grew out of the founders' earlier work on cloud development environments for humans. Daytona initially positioned itself as an open-source manager for standardized developer environments, then shifted toward infrastructure for agent-generated code and long-running agent workflows. In June, Daytona moved its production codebase to closed source, arguing that publicly exposing its isolation layer gave AI-assisted attackers a detailed map of the security boundary they would want to escape.

Burazin's internal push toward agent-heavy engineering therefore serves two purposes. It can reduce the coordination load of hiring, and it gives Daytona's own engineers sustained exposure to the problems Daytona sells customers a way to solve. More internal agents mean more opportunities to encounter failed environments, unclear instructions, state-management problems and review bottlenecks.

The operating model also supports Daytona's commercial argument. Burazin told Business Insider that hiring is "very slow" because every additional employee adds communication overhead. He wants agents to let smaller human groups take on more work without adding the organizational layers that usually accompany growth.

Daytona's capital base gives Burazin room to test that approach. On February 5th, Daytona announced a $24M Series A led by FirstMark Capital, with Pace Capital, Upfront Ventures, Darkmode, E2VC, Datadog and Figma Ventures participating. Daytona said at the time that it had reached a $1M forward revenue run rate in less than three months and doubled that figure six weeks later. Those revenue figures were supplied by Daytona.

Business Insider, citing Crunchbase, reported that Daytona subsequently raised a $48M Series B. Forge separately lists Daytona's latest financing as a July 2026 Series B and puts total funding at $79.07M.

A different hiring filter

If Burazin's experience holds beyond Daytona, the most valuable agent operators will combine functional depth with the communication habits normally developed through management. They will need to break broad goals into bounded tasks, assign the right context, recognize unreliable output and know when to intervene.

That favors experienced builders who can explain their judgment. It also raises the bar for managers whose authority previously came from controlling headcount, meetings or information flow. In an agent-heavy workplace, their work becomes visible in the quality of the instructions, evaluations and decisions surrounding machine output.

Daytona's experience remains an anecdote from a small organization whose business depends on customers running more agents. Burazin is still putting real operating weight behind it. He is keeping human hiring slow, asking engineers to supervise multiple agents and treating management skill as part of the technical stack. The org chart may stay small, but the work of management is spreading into every prompt, specification and review.