PostHog AI Agents Handled 6,124 Error Tracking Issues in 30 Days Without Human Intervention.

PostHog reported that AI agents closed 4,063 errors, suppressed 1,751 known-noise issues, and rerouted 310 problems to appropriate teams over the past 30 days, all without users opening the platform's UI. The numbers come from the company's error tracking Model Context Protocol (MCP) implementation, which PostHog initially designed as a triage assistant but has since realized serves three distinct functions.

The resolution workflow accounted for the majority of automated actions. Agents marked issues as complete after generating a fix, confirming a shipped solution, or tracing root causes to problems that had already self-resolved. This represents a shift from delegating error review to actually closing the loop on bug fixes autonomously.

Suppression emerged as what PostHog called "a quiet win." The 1,751 suppressed issues were filtered as known noise, including common browser errors like "Script error," "ResizeObserver loop completed," and "Object Not Found Matching Id." The company noted that agents are now maintaining noise filters that engineering teams typically intend to manage but rarely prioritize.

The routing function, while representing the smallest portion of activity at 310 issues, may be the most operationally significant. Agents reassigned problems to the appropriate person or team and frequently attached fresh Linear tickets so the new owner had a dedicated workspace to track resolution. This goes beyond simple categorization by creating actionable handoffs with context already attached.

PostHog acknowledged that its original mental model of "the agent does triage" was insufficient. The company now views fix, suppress, and route as three separate jobs requiring distinct prompts and success metrics. Future skills shipped through the MCP will treat each workflow independently rather than bundling them under a single triage umbrella.

The update follows PostHog's recent integration with Synscribe for attribution automation and its Slack agent deployment tool, signaling a continued push toward agent-driven workflows across its analytics platform.

The data suggests that AI agents are moving beyond assistance into autonomous execution for well-defined error tracking tasks. Over 65 percent of the 6,124 issues were fully resolved without human review, indicating that agents can confidently identify fixes, validate deployments, and close tickets when given sufficient context.

Suppression automation addresses a persistent pain point in error tracking: noise management. Teams often know which errors to ignore but lack the time to properly configure filters. Agents handling this maintenance work could reduce alert fatigue and improve signal-to-noise ratios in monitoring dashboards.

Routing with attached Linear tickets introduces a layer of workflow intelligence that standard triage tools lack. By creating tracking artifacts during reassignment, agents ensure that context travels with the issue rather than getting lost in handoff conversations.

PostHog indicated that the next iteration of its MCP skills will differentiate between these three workflows at the prompt level. This means separate agent instructions optimized for resolution validation, noise pattern recognition, and intelligent routing rather than a single general-purpose triage prompt trying to handle all three.

The company has not specified a timeline for shipping the refined skills but noted that treating each workflow independently should improve both accuracy and agent decision-making quality. The 30-day observation period provided enough behavioral data to justify architectural changes in how the MCP handles error tracking automation.