The agent workload is structurally totally different
Brokers are long-running, stateful processes. They motive throughout time, name exterior instruments, spawn subprocesses, write and execute code, and make selections that depend upon what occurred 5 steps earlier in the identical job. A single-agent workflow may run for minutes or hours, touching a dozen exterior techniques and producing intermediate outputs that subsequent steps depend upon. The compute layer for that type of work must do issues the outdated mannequin was by no means requested to do. That’s the new sample: Execution infrastructure designed round agent semantics quite than request semantics.
The Kubernetes group itself has acknowledged this mismatch. In March 2026, Kubernetes SIG Apps revealed an introduction to Agent Sandbox, a brand new CRD-based abstraction designed particularly for singleton, stateful agent workloads. The framing is direct: The ecosystem is transferring from short-lived, remoted duties to deploying a number of, coordinated AI brokers that run repeatedly, and mapping these workloads to conventional Kubernetes primitives requires a wholly new abstraction. The truth that the Kubernetes maintainers constructed a devoted primitive for this, quite than recommending groups compose one from current sources, is itself the clearest sign that agent execution doesn’t match the outdated mannequin.
What agent execution really requires
Concretely, it requires 4 issues. First, remoted execution environments that provision in milliseconds, not minutes, so every agent job will get its personal sandbox for code execution and power calls with out blocking the reasoning loop. The distinction between a two-second surroundings and a two-minute surroundings shouldn’t be a efficiency optimization; it determines whether or not the structure is viable in any respect. Second, sturdy state administration throughout the total job lifecycle, so an agent can pause, hand off or resume with out re-initializing from scratch and burning tokens to reconstruct context it already constructed. Third, coordination primitives for multi-agent work: The flexibility to spawn subagents, move structured outputs between them and observe job dependencies throughout a graph of concurrent processes. Manufacturing agent techniques are hardly ever single brokers; they’re pipelines of specialised brokers with handoffs that should be dependable and inspectable. Fourth, credentials and secrets and techniques administration that journey with the execution context, so brokers can authenticate to exterior companies securely with out exposing credentials within the job definition, logs or the surroundings variables of a shared container.
