Typically, moving into the superb factors of the UI could be painstaking, with a lot of back-and-forth interplay. The AI will repair one factor, then regress one other. At instances, I began to really feel like going into the CSS myself could be faster. That is an space the place the concept of “GUI grounded AI”—that’s, brokers that may visually interface with the UI through laptop imaginative and prescient—shall be worthwhile.
The current and future state of agentic AI
Roo Code paired with Gemini inside VS Code is the simplest setup I’ve discovered but for delivering on agentic promise. That mentioned, the declare that agentic AI will change fairly than increase an precise human programmer remains to be inconceivable. Roo Code is extra like a developer exo go well with than an autonomous robotic.
As a cautious real-world developer—one who’s been bitten by spiraling complexity repeatedly—I’m cautious of the quantity of uninspected code being added to our codebases. A accountable developer will decelerate, flip off the learn/write auto-approval, and look extra deeply on the diffs. They’ll take the time to soak up and perceive the that means of the proposed adjustments and regulate the place wanted.
What’s harmful is the temptation to see AI-generated code that appears kind of okay, and simply hold rolling. That’s okay for the early phases of prototyping, however we nonetheless want skilled builders within the loop in the course of the later phases.
