Tools

Running one AI agent at a time is slow. The real speed gain comes from running several in parallel, each in an isolated branch, and checking in on each as it finishes. These tools make that workflow manageable.

Every other category here helps the agent write code. This one is how it finds out whether the code it wrote actually works, which for Frontend is the only question that matters and the one a type checker cannot answer.

The libraries you reach for when "AI feature" stops meaning "tool inside my IDE" and starts meaning "agent shipping in my customer's product." Not coding assistants — production frameworks for the AI features your team actually ships.

AI agents are stateless by default — they forget everything when the context window resets. These tools give your agent persistent memory across sessions, projects, and teams.