Reflection AI launches Asimov to track full software development lifecycle

Reflection AI, a startup founded by former Google DeepMind researchers, has launched Asimov, an AI agent designed to revolutionize how software development is tracked and understood. Unlike conventional coding assistants, Asimov analyzes not just code but also emails, Slack messages, project reports, and technical documentation to create a comprehensive map of the software development process.

Key Features of Asimov

• Deep-Research Architecture: Uses multiple "retriever" agents to extract relevant details from large codebases, with a "combiner" agent synthesizing the information into coherent responses.
• Asimov Memories: Allows developers to store internal team knowledge with prompts like "@asimov remember X works in Y way," protected by role-based access controls.
• Performance: In internal surveys, developers preferred Asimov’s answers 82% of the time, outperforming competitors like Anthropic’s Claude Code (63%) and Cursor Ask.

Founders and Vision

The startup is led by CEO Misha Laskin (ex-Google DeepMind, contributed to Gemini) and CTO Ioannis Antonoglou (worked on reinforcement learning for AlphaGo). Their goal is to create a persistent memory structure for development teams, capturing decision logs and business context to pave the way for more powerful AI systems.

Deployment and Security

• Runs on open-source models fine-tuned by Reflection using reinforcement learning (no customer data used for training).
• Deployed within customers' virtual private clouds for security.
• Proprietary models are in development for future releases.

Watch Asimov's introduction video

Industry Impact

Reflection’s approach diverges from the industry’s focus on code generation, instead aiming to understand the full development lifecycle. This could lead to more efficient teams and lay the groundwork for advanced AI systems capable of superintelligent software development.

For more AI news, subscribe to THE DECODER Newsletter.