Cognition launches Devin Fusion and Devin Voice AI tools

Cognition has introduced two major updates to its AI software engineering platform, Devin: Devin Fusion and Devin Voice.

Devin Fusion is a multi-model coding architecture designed to improve efficiency by pairing a high-level “lead” model for strategic planning with a more cost-effective “sidekick” model for execution. Cognition reports that this dual-model approach can deliver up to 39% better efficiency and a 46% cost reduction compared to single-model setups. The system utilizes separate persistent contexts for each model to avoid cache misses and allow parallel processing.

Additionally, the company launched Devin Voice, which implements full-duplex voice technology to allow developers to interact with the coding agent through spoken conversation. Utilizing OpenAI’s GPT-Live-1 voice API, the feature enables users to interrupt, redirect, and provide instructions verbally. While the tool offers a more conversational workflow, it has raised questions regarding the ability of development teams to audit and review code changes that are initiated via voice rather than typed commands.