Google's Gemini Code Assist Enhances AI Coding with New Agentic Capabilities

Gemini Code Assist, Google's AI-powered coding companion, is rolling out exciting new "agentic" features in a preview mode. At the recent Cloud Next conference, Google unveiled how Code Assist can now deploy AI "agents" that go the extra mile to tackle complex programming tasks. Imagine these agents taking a product spec from a Google Doc and turning it into a full-blown application, or effortlessly converting code from one language to another. It's pretty impressive stuff, and now you can even use Code Assist within Android Studio, along with other coding environments.

These enhancements come as no surprise, given the stiff competition from the likes of GitHub Copilot, Cursor, and Cognition Labs with their viral tool, Devin. The race to dominate the AI coding assistant market is heating up, and with the potential to boost productivity, there's a gold rush on. The stakes are high, and everyone's eager to get a piece of the action.

Managing Code Assist's Agents

Code Assist's agents are now manageable through a sleek new Gemini Code Assist Kanban board. This board lets you keep tabs on your agents as they generate work plans and report progress on tasks, step by step. These agents aren't just about writing code or migrating it; they're also capable of implementing new app features, running code reviews, and whipping up unit tests and documentation. At least, that's what Google claims.

Performance and Security Concerns

But how well do these agents really perform? That's the million-dollar question. Even the top-tier code-generating AIs today have a tendency to slip in security vulnerabilities and bugs, thanks to their struggles with understanding programming logic. A recent study on Devin, for example, showed it only managed to successfully complete three out of 20 tasks. So, if you're thinking of using Code Assist to create or refactor an app, it might be wise to double-check the code yourself. Better safe than sorry, right?