Honeycomb ships agent timeline

Honeycomb on May 12 introduced Agent Timeline, Canvas Agent and Canvas Skills, expanding its observability platform to cover AI agents operating in production. The San Francisco company said the new products are meant to show engineering teams what agents did, which tools they called, how much they cost and where failures occurred across multi-step workflows. (honeycomb.io) Honeycomb said the tools work without proprietary software development kits or framework lock-in. The launch comes as more software teams move agents into tasks such as code generation, incident triage, cloud operations and customer support. ### What exactly did Honeycomb ship on May 12? Honeycomb said the release includes three pieces: Agent Timeline, a rebuilt Canvas with a Canvas Agent, and Canvas Skills. (honeycomb.io) Agent Timeline is the new investigation view for agent workflows, while Canvas Agent is the company's AI assistant inside its collaborative investigation workspace, and Skills are reusable instructions that encode team-specific workflows and runbooks. The company described Agent Timeline as a conversation-level view that ties together agent invocations, large language model calls, tool calls and downstream traces under a conversation ID. Honeycomb's product page says engineers can inspect prompt details, token usage, API calls, database queries and infrastructure behavior in one place. (honeycomb.io) ### Why build a separate timeline for agents instead of using ordinary tracing? Honeycomb said agent workflows create a visibility problem because they branch, retry, hand off between agents and invoke tools in ways that are not obvious from a single request trace. In a company blog post published May 13, Honeycomb said engineers often end up stitching together fragments by hand because traditional application performance monitoring tools show HTTP requests but not full agent interactions, while model-evaluation tools see model behavior but not database or API calls. (honeycomb.io) Christine Yen, Honeycomb's co-founder and chief executive, said in the launch announcement that AI has introduced "real nondeterminism into production" and changed how humans and agents build and validate software. (honeycomb.io) That framing underpins the product design: the company is pitching observability not just for model outputs, but for the sequence of intermediate decisions and tool use that led to them. ### What does Agent Timeline show that teams could not easily see before? Honeycomb's documentation says Agent Timeline lets users browse recent conversations or open one by conversation ID, then view LLM calls, tool invocations, errors and token usage in chronological order. (honeycomb.io) The product page adds failure counts at the conversation level, red highlights on failing spans and a "Show Failures Only" mode. A May 13 Honeycomb demo post gave a concrete example: a support team passed along a single conversation ID, and the timeline showed that one tool call had failed because a shipping-service connection error had taken a dependency offline. In the same post, Honeycomb said another investigation surfaced an agent repeatedly pulling 145,000 rows of order data and sending them back to the model on every turn, which increased cost and latency. (honeycomb.io) ### Where do Canvas Agent and Skills fit into the workflow? Canvas is Honeycomb's AI-assisted investigation workspace, and the company says the rebuilt version can now operate both as a chat interface and as an autonomous agent. The Canvas product page says the system presents its investigation steps and query results in an interactive notebook so users can review the evidence and continue the analysis with follow-up questions. (honeycomb.io) Agent Skills extend that workflow into coding and debugging tools. Honeycomb's documentation describes Skills as a collection of skills, agents and hooks for AI-assisted code workflows, covering areas such as OpenTelemetry practices, production debugging, SLOs and instrumentation guidance. Honeycomb also maintains a public repository for the package, listing support paths for tools including Claude Code, GitHub Copilot CLI and OpenAI Codex CLI. (honeycomb.io) ### When can customers get the new features? Honeycomb said Canvas, Canvas Agent and Skills will be available starting next week for all Honeycomb customers. Agent Timeline entered early access on May 12 and is expected to reach general availability next month, according to the company's launch announcement and press release. (honeycomb.io) Honeycomb is also using the launch in its Innovation Week programming. The company posted product videos on May 13 showing Agent Timeline and the reworked Canvas in action, and its newsroom says O11yCon 2026 is scheduled for May in San Francisco with speakers from Google, Salesforce, Stack Overflow and Superhuman. (honeycomb.io 1) (honeycomb.io 2) (docs.honeycomb.io)