AWS just released its Cursor killer…

Amazon’s new AI coding IDE, Kira, is positioned as a “Cursor killer” by tackling a core weakness in many coding assistants: the tendency to generate messy, unstructured output. Instead of jumping straight from prompts to code, Kira pushes users through a spec-driven workflow—starting with a requirements document (user story and acceptance conditions), moving to a design document (component structure, testing strategy, and error handling), and only then producing an implementation plan that breaks work into tasks the AI can execute. The result is slower than typical “vibe coding,” but more controlled—an approach aimed at teams and enterprise environments where consistency, reviewability, and quality matter.

Early hands-on impressions point to a polished interface, but also highlight growing pains. The IDE felt sluggish at times, lacked certain features users expect from established VS Code forks (including chat checkpoints and some Cursor-like capabilities), and ran into delays consistent with an overloaded early release service. Still, the workflow itself stood out as the biggest differentiator: Kira tries to prevent “slop” by forcing structure before code generation. Users click a button to begin tasks once the requirements and design documents are combined into an implementation plan, turning vague intent into a staged engineering process.

The competitive context is intense. The transcript frames a market where major companies pay billions for developer mindshare through VS Code forks—citing the failed OpenAI acquisition of Windsurf, Google’s intervention that reportedly gutted Windsurf talent, and Cognition’s subsequent purchase of Devon-related “scraps.” Against that backdrop, Amazon’s timing is treated as strategic: Kira is powered by Claude Sonnet 4.0 and is currently free, with claims that specdriven development will handle complexity better than “sloppy” alternatives.

Pricing and model access are also central to the rivalry. Cursor is described as effectively constrained by Anthropic’s economics: if developers want Claude, Cursor must act as a middleman and price accordingly. The transcript notes that Cursor previously adjusted pricing in a way that increased customers’ bills, prompting a public apology. Kira’s initial pricing—paired with its Claude-only setup—could therefore undercut Cursor by delivering more Claude utility at lower cost, at least until other models are added.

Finally, the transcript widens the lens beyond Amazon. It mentions China’s release of Kimmy K2, an open-weight agentic coding model said to approach Claude’s performance. If multiple strong models converge with IDEs that can manage complexity, the “AI IDE arms race” could shift from raw code generation to disciplined engineering workflows. For now, Kira’s promise hinges on whether its structured approach and enterprise focus outweigh its early-release friction—and whether it can sustain momentum once the service is no longer overloaded and feature gaps close.