Dust vs Kubernetes
Side-by-side trajectory, velocity, and editorial themes.
Dust is widening the agent-platform surface: multimodal tools, enterprise audit, model breadth.
Dust is shipping at a fast clip on three fronts that together define a serious agent platform: model breadth (Gemini 3.5 Flash, Grok 4.3, refreshed Anthropic lineup), agent capability (MCP tools can now return images the agent can actually see, plus context compaction for long runs), and enterprise readiness (workspace audit logs streamable to Datadog, Splunk, or any HTTPS sink). Integrations are getting versioned upgrades on the side (Asana MCP v2, Gmail labels and archive). The product is moving from 'chat with an agent' toward 'run agents in production with observability and multimodal I/O.'
Two clear directions: deeper enterprise GTM via SIEM-grade audit, and a more capable agent runtime that can see, remember, and act inside third-party SaaS. The MCP-image release in particular treats Model Context Protocol as a real I/O surface rather than a text-only RPC, which is where the broader MCP ecosystem is heading. Frequent model rotations suggest Dust is positioning as model-agnostic infrastructure rather than locking into one provider.
Next moves likely lean into the same arc: more MCP integrations with action verbs (write/delete/transition states), expanded multimodal returns (audio, structured documents), and finer-grained admin controls layered on top of the audit foundation - tool-usage policies, per-agent egress rules, or approval workflows.
Kubernetes 1.36 leans into AI/ML scheduling and control-plane scaling.
The 1.36 cycle is graduation-heavy, with PSI metrics, declarative validation, and volume group snapshots all promoted to GA. Alongside that, the project is making architectural moves around workload scheduling (a new PodGroup API), API-server safety (Mixed Version Proxy on by default), and very-large-cluster scaling (server-side sharded list and watch in alpha). Etcd 3.7 has hit beta in parallel.
Kubernetes is repositioning the control plane for two pressures at once: AI/ML batch workloads, where gang scheduling and DRA are becoming first-class concerns, and very-large clusters, where the control plane itself needs to shard. The pattern across this cycle is consolidation — old experimental scaffolding is reaching GA or being removed (ExternalIPs), while new APIs land with explicit separation of static template from runtime state. Less feature sprawl, more API hygiene.
Expect 1.37 to push server-side sharded watch toward beta and to keep extending DRA's reach into native resources like memory and networking. Workload-aware scheduling will likely accumulate scheduler-plugin-level coordination patterns next, with downstream batch frameworks starting to converge on the PodGroup shape.
See more alternatives to Dust →
See more alternatives to Kubernetes →