JetBrains has shipped a significant upgrade to GoLand 2026.3, integrating its built-in profiler directly with AI coding agents. The update allows agents like Claude Code, Codex, and Junie to autonomously run applications, capture performance data, and identify bottlenecks without requiring the developer to manually interpret low-level pprof output. This shift moves performance debugging from a specialized task to an automated workflow, leveraging Model Context Protocol (MCP) tools to bridge the gap between AI reasoning and IDE execution.

Automating the Rescue Mission

Historically, Go developers avoided pprof because it felt like a rescue mission for unfamiliar territory. The new goland-profiler skill changes this dynamic by letting the agent handle the mechanics. When a user reports a symptom, such as slow reindexing, the agent initiates the profiling run, selects the appropriate profile type (CPU, block, mutex, or heap), and analyzes the results. It returns a summary of the hottest functions and call paths, effectively translating raw data into actionable insights while keeping the source code untouched.

Three Modes of Agent Collaboration

The integration supports three distinct workflows to accommodate different developer preferences. In the 'Agent does everything' mode, the AI handles the entire lifecycle from launch to analysis. For those who prefer control, the 'You profile, agent analyzes' mode allows developers to capture profiles manually and have the AI interpret the open flame graph. A hybrid approach lets developers manage the runtime state while the agent captures specific metrics, ensuring the context is precise before analysis begins.

Technical Constraints and Limitations

Despite the automation, the tool has clear boundaries. GoLand’s profiler modifies the executable during compilation to expose the API, meaning it cannot attach to already running processes. Additionally, the sampling mechanism struggles with short-lived CLI tools that finish in fractions of a second. The feature is currently verified only for Claude Code, Codex, and Junie, though other MCP-compatible agents may function. Developers must ensure the goland-profiler skill is enabled in Settings | AI | Skills to activate the integration.

Key Takeaways

  • GoLand 2026.3 enables AI agents to operate the built-in profiler via MCP tools.
  • Supported agents include Claude Code, Codex, and Junie.
  • The agent automatically selects profile types (CPU, mutex, heap) based on reported symptoms.
  • Limitations include inability to attach to running processes and poor support for short-lived CLI tools.

The Bottom Line

This is a pragmatic win for AI-assisted development, moving agents from code generation to system-level diagnostics. By offloading the cognitive load of pprof interpretation, JetBrains makes performance tuning accessible to agents, potentially halving the token usage for debugging tasks.