Agent workloads create unusual pod churn
Google Cloud has described how GKE Dataplane V2 supports very large fleets of short-lived AI-agent sandboxes. Reinforcement-learning and code-execution workloads can create and destroy thousands of isolated environments each minute. At that scale, slow sandbox startup leaves expensive GPUs and TPUs waiting, turning networking behaviour into a material part of model-training economics.
GKE Dataplane V2 uses eBPF and maps pod labels to numeric network identities for policy enforcement. Agent frameworks often attach unique labels such as agent identifiers, timestamps and claim references to every pod. Rapidly allocating a new identity for each sandbox can pause creation and approach Cilium’s cluster-wide limit of 65,536 identities.
High-cardinality labels exhaust network identities
Google recommends moving high-cardinality information that is not used for service selection or resource grouping from labels to annotations. Operators can also configure label-exclusion filters so dynamic framework metadata is ignored during identity calculation. Identical agent types can then share an identity while retaining the labels required for functional policy decisions.
Google says this approach supports 300 sandboxes per second with sub-second latency. Its higher-density Agent Substrate builds on the same primitives while bypassing Kubernetes scheduling limits. In a benchmark with 200,000 agents, the company reports 500 activations per second and suspend or resume latency below 500 milliseconds.
Agent Substrate pushes beyond scheduler limits
Label filtering must be precise because a field excluded from identity calculation can no longer distinguish network policy. Teams should classify every label by whether it is decorative, operational or security-relevant, then test that agents remain isolated across tenants, tasks and privilege levels. Performance should not be purchased by merging identities that require different access.
The design is useful for organisations whose agent economics depend on extreme concurrency, but most deployments should first measure their actual bottleneck. A simpler cluster may be easier to govern at modest scale. When density is justified, capacity testing should include control-plane failure, noisy neighbours, policy updates and abandoned sandboxes, not only best-case activation throughput.
What teams should validate before production
For implementation, teams should record the exact model, software, dataset, connector and policy versions behind every result. Representative evaluation must include normal work, boundary cases, deliberate misuse and component failure, with outcomes separated by user role, data sensitivity and consequence. A strong average should never hide a serious failure in a critical path. Changes should pass the same regression set before promotion, and operators need a documented fallback that preserves evidence rather than silently broadening access or lowering quality.
Procurement and planning should account for the human work required to review exceptions, maintain controls and investigate incidents. Published performance or efficiency claims are useful starting points, but production confidence comes from evidence collected in the organisation’s own environment. A named owner, measurable acceptance threshold, cost model and tested shutdown path turn a promising release into a service that can be governed over time. Where decisions affect safety, rights, money or regulated records, accountable professional review remains essential.
Security and operational evidence
Security review should map every identity, credential, network destination, stored artefact and administrative access path involved in the workflow. Logs need enough context to reconstruct who initiated an action, which evidence was consulted, what the system proposed and what ultimately executed. Sensitive prompts and traces should not be retained merely because they are useful for debugging. Retention, redaction and access must follow the underlying data classification, with a practical process for correcting or removing material when policy or law requires it.
Operational monitoring should measure quality and service behaviour together. Latency, throughput and availability matter, but so do unsupported claims, missed constraints, unsafe tool calls and the rate at which people overturn recommendations. Teams should rehearse dependency outages and partial failures before unattended operation. If a model, connector or external service becomes unavailable, the workflow should fail predictably, preserve its state and tell the user what remains unfinished. Clear degradation is safer than an agent improvising around a missing control.
Transparency should scale with adoption
Organisations should explain the scope and limitations of the deployed capability to the people affected by it. Users need to know when automation is involved, how to challenge an outcome and where a human can intervene. That transparency improves feedback and prevents a pilot result from being interpreted as blanket approval for broader autonomy. Expansion should follow demonstrated evidence, not novelty or pressure to match a competitor announcement. The durable advantage comes from dependable integration, careful measurement and learning that survives the next model or product update.
Finally, teams should schedule a review after real usage rather than treating launch as the end of governance. The review should compare expected benefits with observed quality, cost, incidents and user behaviour, then retire workflows that do not justify their complexity. A service that works well can expand through the same evidence-based process. A service that fails should be narrowed or stopped without losing the records needed to understand why. This feedback loop is what turns experimentation into accountable operational capability.