newsfilter
← Latest News
Sequoia Capital·

Reflection AI Launches Asimov: Breakthrough Agent for Code Comprehension

Summary

  • Reflection AI launched Asimov, a research agent designed for code comprehension & understanding.
  • The product targets the majority of engineering time spent on parsing complex codebases rather than writing code.
  • In blind testing against Cursor Ask & Claude Code (Sonnet 3.7 & 4), Asimov answers were preferred by maintainers of large open-source projects.
  • Asimov builds a single source of truth by ingesting codebases, architecture docs, GitHub threads, & chat history.
  • The tool features "Asimov Memories" to capture team-wide tribal knowledge with permissioned role-based access control.
  • Engineers can update Asimov's knowledge directly to offload context from senior staff to the entire team.
  • The architecture uses multi-agents with small long-context retrievers & a large short-context reasoning agent for synthesis.
  • The current release utilizes third-party models while Reflection AI actively trains its own models for future performance.
  • Reflection AI applies a strategy coupling model intelligence with end-application development for superintelligence.
  • Asimov serves as the first step in Reflection AI's path toward building powerful coding agents.

Asimov, the breakthrough research agent for code comprehension that accelerates engineering teams, is the first product launch from Reflection AI.