Copilot shifts, Gemini hardens: 4 major AI coding changes

Four major change groups qualified in the 24-hour window ending at 20:15 (UTC+6). The practical split is simple: update Gemini CLI, plan the Copilot MAI model migration, and review whether PyCharm's new agent skills fit your Python workflow.

GitHub Copilot is rolling out MAI-Code-1.1-Flash with native image understanding and a 0.25× premium-request multiplier. Business and Enterprise admins must enable it; MAI-Code-1-Flash retires across Copilot on 10 September 2026. 12 Copilot for JetBrains also adds persistent memory, Ollama local models, and more managed settings. 3

Gemini CLI v0.55.1 is the update-now item. It hardens the A2A server against remote-code execution, closes path and credential leaks, and stops capacity-exhaustion retry hangs. 4 PyCharm 2026.2.1 gives agents a live Jupyter kernel and the correct project interpreter; both skills require a JetBrains AI subscription. 56 PyCharm also unbundles and deprecates its Data Wrangler, Hugging Face, and Google Colab plugins. 7

Smaller changes: Claude Code v2.1.228 fixes a Remote Control /resume history leak and hardens synced skills. 8 Copilot usage reports now break token counts down by model. 9 Codex published three note-less alpha builds; stable remains 0.147.0. 10 No product release qualified for Cursor or Windsurf on the first-party web surfaces checked. 1112 Windsurf's X timeline was unavailable, and JetBrains still lacks a unified plugin-level AI changelog.