Rider 2026.2 Released — Agents Can Now Understand Your Code's Execution Performance

The era of AI simply reading and fixing code files is over. If previous AI assistants were like mechanics looking only at a repair manual, we have now entered a time where the AI can start the engine, listen to the sound, and diagnose problems directly.

The recently released JetBrains Rider 2026.2 gives AI full access to compilation logs, test results, and even precise performance analysis data. Beyond just skimming through code as text, collaboration begins with a smart agent that pinpoints bottlenecks by observing actual execution status.

An Agent That Goes Beyond Code Scanning to View Execution Status

Existing AI coding assistants were limited to reading and analyzing only source code text. Without knowing how the code actually runs or where bottlenecks occur, they often relied on mere guesswork to fix issues. JetBrains Rider 2026.2 completely shatters this barrier.

The core of this update is using the Model Context Protocol (MCP) and Agent Client Protocol (ACP) to feed real-time execution data from the IDE directly to the agent. Notably, the finding-tests feature, which maps data from the test coverage tool dotCover, allows the AI to extract only the necessary execution context without needing to scan every file, reducing token consumption by up to 50%.

Coupled with the dottrace-analyze feature, which analyzes snapshots from the performance tool dotTrace, the agent's diagnostic capabilities evolve to another level. In practice, the AI's accuracy in identifying performance bottlenecks using precise analytical data jumped dramatically from 4.71 to 8.15 out of 10. Coding agents can now perform sophisticated troubleshooting by looking through the flow of execution, moving well beyond static code analysis.

Unexpected Hurdles When Integrating with Strict Clients

No matter how powerful a tool is, there are always growing pains. While the new features in Rider 2026.2 are very exciting, being an initial release, you may encounter some unexpected errors during integration—especially when working with clients that strictly follow protocol specifications.

A representative issue is the IJPL-230494 bug, a schema mismatch. When processing data within Rider, it should explicitly include 'false' values in the response but omits them. Clients that validate formats strictly, like Claude Code, immediately throw an MCP error -32602 and stop working because they detect missing required data.

The 'communication mismatch' during initial setup is another factor frustrating developers. Ideally, the components should perform a formal handshake to begin the connection, but Rider's execution tool fires off ready notifications prematurely, causing the sequence to get tangled. Consequently, the client waiting for a standard response ends up timing out and disconnecting.

These issues are common friction points in a technological transition. For a smart execution environment and a sophisticated agent to work in perfect sync, we need a little more time to observe and refine these initial specification adjustments.

The Beginning of Agent Collaboration Sharing Execution Context

Rider 2026.2's latest effort clearly demonstrates that development tools are evolving from simple code editors into collaborative spaces where the agent and the environment perfectly share execution context. While minor bugs in the initial integration need to be addressed, the move to drastically reduce unnecessary token waste by sharing execution data is a very clear and positive direction.

If this approach becomes the standard, the effort developers spend manually copying and pasting dozens of lines of logs to find the root cause of a bug will be significantly reduced. It will be an interesting point to watch how quickly this workflow—where the agent automatically runs tests and precision analysis tools to provide solutions—integrates reliably into actual development environments.