Former Google Researchers Develop New AI Agent for Software Assistance

A new kind of artificial intelligence agent, named Asimov, has been developed by Reflection, a startup founded by former top AI researchers from Google. Asimov is trained to understand how software is built by analyzing a company’s data, including code, emails, messages, and project updates. The goal is to learn how these components come together to create a finished software product.

Reflection aims to build superintelligent AI, with the CEO Misha Laskin emphasizing the importance of AI mastering coding to interact with the world effectively. Asimov focuses on reading code rather than writing it, aiming to be a useful coding assistant in team settings.

Asimov is composed of multiple smaller agents working together to understand and respond to user queries about code. Reflection claims that Asimov outperforms some leading AI tools, with developers preferring its answers in a survey conducted by the company.

While the approach seems promising to some experts, such as computer scientist Daniel Jackson from MIT, concerns remain about potential computation costs and security issues, especially related to privacy. Asimov operates within customers' virtual private clouds to ensure data retention by the customer.

According to the source: WIRED.