We Gave AI Agents a Live Jupyter Kernel in PyCharm - The JetBrains Blog

PyCharm’s new Jupyter skill lets AI agents work directly with a live Jupyter kernel rather than editing notebook JSON files and launching disposable shell subprocesses, preserving variables, models, imports, and data across notebook cells while reducing corruption risks. Through an MCP-based tool interface, agents can create, edit, and run notebooks, monitor kernels, wait for lengthy executions without repeated polling, and retrieve only newly streamed output, reducing idle context usage. In tests on 12 MLGym machine-learning tasks, Claude Opus 5 completed all tasks in kernel, shell, and mixed modes, but the kernel-only approach cost $59.09 versus $67.06 for shell-only operation, largely because 98% of inputs were lower-cost cache reads compared with 82% for the shell. Cost advantages varied by model and workload, with the kernel especially beneficial for long stateful Opus tasks but not always cheaper for short tasks or Codex models. The feature, available through JetBrains AI subscriptions in PyCharm 2026.2.1, improves notebook workflows but still requires agents to be instructed to save artifacts and does not overcome difficult machine-learning problems that need stronger methods or human involvement.

No tracked trend matches for this post yet.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.