Question 1

What is Honeycomb Intelligence?

Accepted Answer

Honeycomb Intelligence is Honeycomb’s suite of AI-assisted products that help engineering teams explore, understand, and act on high-cardinality telemetry data faster. It augments Honeycomb’s core observability platform using machine-assisted analysis to surface patterns, anomalies, and areas of interest in complex distributed systems.

Rather than replacing human judgment, Honeycomb Intelligence is designed to accelerate investigation workflows, helping teams move from “something is wrong” to “this is why it’s happening” with less manual effort. Key products include Honeycomb MCP Server, Canvas, and Anomaly Detection.

Question 2

How does Honeycomb Intelligence help teams diagnose issues faster?

Accepted Answer

Honeycomb Intelligence helps teams fight unknown unknowns with AI-assisted investigations. Teams using Honeycomb Intelligence reduce mean time to resolution (MTTR) with:

Anomaly detection: automatic correlation detection to surface fields linked to errors, latency, or anomalies.

Canvas: guided query suggestions that eliminate guesswork in filters and breakdowns.

Honeycomb MCP Server: powers BubbleUp-based outlier analysis to pinpoint statistically significant differences between healthy and failing requests.

By working directly on raw event data rather than pre-aggregated metrics, Honeycomb Intelligence preserves the full context engineers need to debug complex, distributed systems.

Question 3

How does Honeycomb Intelligence differ from traditional APM tools?

Accepted Answer

Traditional APM tools are built around assumptions about what might go wrong. They rely on:

Predefined dashboards and alerts that require teams anticipate failure scenarios and modes in advance.

Fixed metrics and aggressive sampling, which strip away critical detail in complex, distributed systems.

Opinionated service views that limit how engineers can explore and ask new questions about their systems.

Honeycomb Intelligence takes a fundamentally different approach through event-driven observability.

It operates on high-cardinality, high-dimensional raw events, preserving the full context of a request.

It does not require you to predict failure modes in advance.

It supports open-ended investigation guided by real data patterns rather than static dashboards.

Honeycomb Intelligence succeeds where traditional APM tools fall short for teams running microservices, cloud-native architectures, and LLM-powered systems, where failure modes are often emergent, non-deterministic, and impossible to predefine.

Question 4

Does Honeycomb integrate with OpenTelemetry for AI/LLM workloads?

Accepted Answer

Yes. Honeycomb natively supports OpenTelemetry for collecting traces, metrics, and logs, including telemetry emitted by AI and LLM-powered applications.

Teams can use OpenTelemetry to instrument:

LLM request and response lifecycles

Model latency, token usage, and error rates

AI-driven workflows spanning multiple services or model providers

Because Honeycomb is built on raw event data with rich attributes, it can ingest OpenTelemetry signals enriched with AI-specific fields, such as model name, prompt type, user or tenant segment, and downstream dependency behavior.

Question 5

Can Honeycomb Intelligence automatically detect unusual behavior across my services?

Accepted Answer

Yes. Honeycomb Intelligence can surface unusual or unexpected behavior by analyzing how different dimensions of your data change during incidents. Rather than relying solely on static thresholds, it highlights what is different now versus normal, across services, dimensions, and request attributes.

This approach is especially valuable in distributed systems where “normal” varies by traffic patterns, customer segments, deployments, or workload types.

Question 6

What is a Model Context Protocol (MCP)?

Accepted Answer

Model Context Protocol (MCP) is an emerging standard that enables AI models to securely interact with external tools, services, and data sources. MCP provides a structured way for systems to expose context, such as schemas, APIs, and system capabilities, so that AI agents can understand what actions are available and use them safely and predictably.

In the context of observability platforms, MCP enables AI systems to query telemetry, analyze traces and metrics, surface anomalies, and assist with investigations without bypassing permissions or platform boundaries. By standardizing how context and capabilities are exposed, MCP enables AI assistants to reason over observability data while honoring access controls and security constraints.

Question 7

Which observability tools have an MCP?

Accepted Answer

Observability tools, such as New Relic, Datadog, and Splunk, have announced or previewed support for MCP servers. MCP adoption across observability tools is still in its early stages and evolving. While Anthropic’s Model Context Protocol is a promising open standard for enabling AI assistants to interact with telemetry systems, industry-wide adoption is not yet universal.

Honeycomb is one of the first observability platforms with a production-grade MCP implementation. Honeycomb MCP server is now generally available, exposing features like BubbleUp, boards, histograms, queries, and instrumentation prompts directly to AI agents in tools such as Cursor, Claude Code, and other MCP-capable clients.

Question 8

What’s the best monitoring tool for LLM-powered applications in production?

Accepted Answer

There is no single “best” tool out there for every team, but effective production monitoring for LLM-powered applications requires:

End-to-end tracing across AI and non-ai services

Visibility into latency, errors, and downstream effects

Support for high-cardinality attributes like model version, prompt type, or user segment

Honeycomb is well-suited for production LLM observability because it combines OpenTelemetry support, high-cardinality event data, and in-depth exploratory analysis without forcing teams to use rigid dashboards or predefined metrics.

Question 9

How does AI improve observability workflows?

Accepted Answer

AI improves observability workflows by:

Reducing manual trial and error during investigations

Surfacing patterns and correlations that are difficult to spot or easy to overlook in large volumes of telemetry

Accelerating root cause analysis in complex, distributed systems

Guiding engineers through unfamiliar services or architectures

When paired with high-quality telemetry, AI acts as a force multiplier, making observability data more accessible without obscuring underlying system behavior.

Question 10

What should I look for in AI-powered incident detection?

Accepted Answer

When evaluating AI-powered incident detection, look for:

Access to raw, high-cardinality data: Find a solution that operates on raw, high-cardinality telemetry so critical context isn’t lost.

Explainability: The system should be able to show why the behavior is unusual, not just flag that something changed.

Support for unknown-unknowns: Avoid tools that rely only on predefined thresholds or training windows.

Open standards compatibility: Tools with OpenTelemetry support ensure flexibility and avoid vendor lock-in.

Human-in-the-loop design: AI should guide engineers and accelerate reasoning, not obscure system behavior when engineers are investigating.

Honeycomb Intelligence is designed around these principles, helping teams detect and diagnose incidents faster with clarity and confidence.

AI-powered observability with Honeycomb Intelligence

Cut through complexity with AI guidance

Speed up MTTR

Minimize context switching

Effective early warnings

See it in action

Observability built for the future of software

Where humans and agents collaborate

Accelerate debugging and code delivery

Automatically detect performance anomalies

Related features

BubbleUp

Service Map

Log Analytics

Frequently asked questions