New Capabilities and Limitations of AI Coding Assistants

New Capabilities and Limitations of AI Coding Assistants

Recent updates to popular AI coding assistants like Claude Code, Codex, and Cursor have introduced several new tools and functional changes. However, alongside these improvements, certain limitations remain. We take an honest look at what is actually worth your attention.

Claude Code: Parallel Tasks and Git Worktrees

The new version of Claude Code implements support for parallel tasks using Git worktrees. This allows you to run multiple tasks simultaneously without edit conflicts. However, it is worth noting that the model itself has not been updated, and the improvements are limited to workflow organization rather than expanding the model's functionality.

Codex: Managing Git Worktrees

Codex has introduced tools for managing Git worktrees, which can significantly improve isolation and parallel workflows. Nevertheless, these changes concern architecture rather than the AI's own capabilities. The primary goal is to ensure safety and prevent conflicts, not to increase intellectual capacity.

Cursor: Data Deletion Restrictions

The Cursor CLI has introduced protection against potentially destructive deletion commands. This is an important step for ensuring safety and control, but it does not add new power for code creation or analysis. It should be remembered that the primary focus is on security, not on expanding functionality.

General Limitations

Overall, the updates focus on process isolation and security, which are undoubtedly necessary for the stable operation of these tools. However, none of the updates make the models smarter or more capable—these are, rather, improvements in management and interaction.

If you want to test the capabilities of these models in real-world tasks, try them out in GEMERA AI, where they are available to users of all skill levels.