Dust vs Honeycomb
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.
Honeycomb is rebuilding observability around an autonomous investigation surface called Canvas.
Every meaningful release in the last quarter rolls up to one product motion: Canvas, an agentic investigation surface that Honeycomb is propagating across the entire product. The May 20 launch turned Canvas into a multiplayer workspace where humans and AI agents investigate together, with auto-investigations that kick off when triggers fire, GitHub-grounded analysis, custom skills for runbook knowledge, and a Slack app. Around the headline launch, Honeycomb shipped BubbleUp Insights (AI-summarized anomaly diffs), a Gen-AI tab in trace view, Query Math, dark mode, and earlier beta surfaces of Ask Canvas and Slack Canvas that the big release now consolidates.
Honeycomb is repositioning from 'query your telemetry' to 'investigate with agents that know your system.' Canvas is the through-line: it shows up on Home, in Slack, in alert flows, in traces. The Gen-AI trace tab and BubbleUp Insights point at a parallel bet - that the kind of system worth observing increasingly includes LLM-powered apps, and the observability tool has to speak that language natively. Together this is a category-redefining move on the AI-native ops front, where competitors are still bolting chatbots onto dashboards.
Expect Canvas to keep absorbing surface area: deeper IDE/GitHub integration so investigations can suggest or open PRs, marketplace-style sharing of custom skills, and Canvas access via MCP so agents in other tools can query Honeycomb directly. The next spark will likely be Canvas writing back to the system - e.g., proposing config changes or runbook edits from what it learned.
See more alternatives to Dust →
See more alternatives to Honeycomb →