Observability company Honeycomb is expanding into AI site reliability engineering (AI SRE), as enterprises deploy increasingly complex AI agents and struggle to understand their behavior in production. The company reports that its annual recurring revenue grew more than 70% year over year and that usage of its Model Context Protocol (MCP) integration increased sharply, with AI agents and MCP-driven workflows now generating more than 4.5 million queries per month. Its latest product push includes Team Agents, designed to help engineering teams monitor deployments, investigate incidents, improve instrumentation and assess agent quality. The strategy reflects a shift in observability from collecting telemetry to using AI agents to investigate and manage software systems.
AI Agents Are Creating a New Observability Problem
The growth of AI agents is changing the operational demands placed on software engineering teams. Traditional applications generally follow predictable execution paths, making it possible to investigate failures through established logs, metrics and traces. Agentic systems are less predictable. They can select different tools, make multiple model calls, retry operations, hand tasks between agents and produce different outcomes from similar inputs.
That makes it harder to determine why a system behaved unexpectedly, where a failure originated or whether an agent’s actions created downstream problems.
Honeycomb is positioning its observability platform around this challenge. Its strategy combines high-cardinality telemetry, AI-assisted investigations and autonomous SRE agents that can monitor production systems and work toward goals defined by engineering teams.
The company reported that annual recurring revenue increased more than 70% year over year, its AI-native customer count grew 130%, and net revenue retention reached 125%. It also said more than 70% of its customer base uses Canvas or its MCP integration weekly. These are company-reported performance figures rather than independently audited market measurements.
MCP and Canvas Bring Production Context to AI
Honeycomb’s earlier investments in Canvas and its Model Context Protocol integration provide the foundation for its AI SRE expansion.
Canvas is a collaborative workspace where engineers can investigate production issues using telemetry and supporting context. The MCP integration allows compatible AI agents and tools to query Honeycomb’s observability data, giving them access to information about system behavior rather than relying exclusively on user prompts or summarized dashboards.
The company says MCP-driven queries grew more than 200 times between January and October 2026. AI agents and MCP now generate more than 4.5 million queries monthly against Honeycomb, compared with approximately 20,000 in January.
The distinction is important for agentic operations. An AI assistant that can access traces, logs and metrics has more evidence for diagnosing a production issue than one working from a short incident description. Honeycomb’s emphasis on high-cardinality data is intended to preserve detailed context, including the relationships between requests, services and individual events.
This can help engineers investigate problems involving multiple services or AI model interactions without losing the details that explain an unusual outcome.
Team Agents Move From Investigation to Continuous SRE Work
Honeycomb’s Team Agents represent the next stage of the strategy. Rather than waiting for an engineer to ask a question or an alert to trigger an investigation, these agents can operate against goals established by engineering teams.
Examples include monitoring software releases, improving instrumentation, investigating alerts, assessing AI agent quality and maintaining technical documentation. Teams can configure agents to run on schedules, triggers or service-level objectives.
The agents build shared memory from previous investigations and connect findings to the underlying queries, traces, deployments or tickets. Honeycomb says they distinguish confirmed findings from hypotheses, giving engineers evidence to evaluate rather than presenting every generated conclusion as established fact.
Actions are also controlled through team-defined permissions. Depending on the authority granted, an agent may create a ticket, notify an incident responder or draft a fix. The company says teams decide how much autonomy each agent receives.
This is a significant distinction from conventional incident-response automation. The goal is to identify reliability problems before they become customer-facing incidents, including regressions introduced during deployment and instrumentation gaps that make systems harder to diagnose.
Honeycomb’s Personal and Team Agents were announced in early access in November 2026. Their ability to operate reliably across different engineering environments will be an important test of the company’s AI SRE proposition.
AI Ecosystem Adds Fleet-Level Visibility
Beyond SRE automation, Honeycomb is expanding its observability capabilities to cover fleets of AI agents.
Its AI Ecosystem provides visibility into agent performance, behavior and costs, while allowing teams to investigate individual conversations. This matters as organizations move from experimenting with a single agent to deploying multiple agents across customer service, software development and internal operations.
Monitoring only model latency or token consumption may not explain why an agent is failing. Engineers also need to understand tool calls, handoffs, execution paths and the application services involved. Honeycomb’s September 2026 product updates introduced additional MCP connectors intended to bring source code, incidents, runbooks and tickets into investigations.
The company has also introduced anomaly detection and Honeycomb Private Cloud, offering dedicated observability environments with managed and self-managed options. These capabilities address operational requirements around system monitoring, data control and enterprise deployment.
Observability Becomes Part of the AI Infrastructure Stack
The market opportunity is growing alongside AI deployment. Gartner forecasts worldwide AI spending of approximately $2.7 trillion in 2026, with AI infrastructure accounting for the largest share of that investment. Its September forecast also projects $461.6 billion in AI software spending. Gartner
As organizations deploy more models and agents, observability becomes necessary not only for traditional application performance but also for AI-specific reliability, cost and behavioral analysis.
Honeycomb is competing in a market that includes established observability providers such as Datadog, Dynatrace, New Relic and Splunk, alongside cloud-native monitoring services from AWS, Microsoft Azure and Google Cloud. Its differentiation rests on detailed telemetry and the use of AI agents to investigate and improve production systems.
The company is also expanding partnerships with AWS and Anthropic, among others, and participated in the public preview of OpenTelemetry tracing for the Claude API and Claude SDK. OpenTelemetry’s role is important because shared instrumentation standards can help organizations collect and correlate telemetry across different services and AI providers.
For enterprise buyers, the emerging requirement is broader than monitoring whether an AI application is online. They need to understand what agents are doing, what resources they consume, how their decisions affect other systems and whether their actions remain within operational boundaries.
Honeycomb’s expansion into AI SRE reflects that shift. Its longer-term opportunity will depend on whether its agents can consistently deliver useful investigations and safe operational improvements while preserving the evidence and human oversight that production engineering requires.
Market Landscape
AI observability is expanding from conventional application performance monitoring into the management of agent behavior, model interactions and autonomous operations.
- Agent observability: Enterprises need visibility into agent decisions, tool calls, handoffs, execution paths and outcomes.
- AI SRE: Vendors are introducing agents that monitor deployments, investigate incidents and identify reliability problems proactively.
- High-cardinality telemetry: Detailed traces, logs and metrics help explain complex system behavior that aggregated dashboards can miss.
- AI cost management: Organizations need to track model usage, token consumption, latency and the operating costs of agent fleets.
- Open standards: OpenTelemetry and MCP integrations are becoming important for connecting AI agents with enterprise telemetry and engineering tools.
Honeycomb’s strategy puts it in competition with Datadog, Dynatrace, New Relic, Splunk and hyperscaler monitoring services. Its focus is on combining detailed production context with AI-driven investigation and goal-oriented SRE automation.
Top Insights
- Honeycomb reported more than 70% year-over-year ARR growth and a 130% increase in its AI-native customer count.
- MCP-driven queries increased more than 200 times between January and October, reaching over 4.5 million monthly queries.
- Team Agents can monitor deployments, investigate alerts, improve instrumentation and assess AI agent quality using defined goals.
- Honeycomb’s AI Ecosystem provides visibility into agent behavior, performance and costs across deployed agent fleets.
- AI SRE adoption will depend on reliable evidence, controlled agent permissions and measurable improvements in production reliability.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
