Reflection just launched Asimov, a new AI agent built to master software development by digesting company data—code, emails, Slack chats, project docs—to understand how software is actually made.
Unlike the usual AI focused on just writing code, Asimov spends more time reading code and related info to become a better teammate for developers. It’s designed as a team player, not just a code generator.
Asimov operates as a group of smaller agents inside one larger one, pulling info and synthesizing clear answers for user questions. Early feedback claims it beats Anthropic’s Claude Code Sonnet 4 in dev preferences: 82% to 63%, according to Reflection’s own survey. 
Reflection is a startup co-founded by ex-Google AI researchers aiming to build superintelligent AI. CEO Misha Laskin says mastering coding is the most natural way to make smarter AI agents interact with the world.
This comes as big players like Meta launch superintelligence projects with massive funding.
There are concerns, though. MIT’s Daniel Jackson points out the approach could spike compute costs and create security risks by reading private messages.
"It would be reading all these private messages,"
Daniel Jackson, MIT
Reflection says Asimov runs inside customers’ virtual private clouds, keeping all data under customer control.
Keep an eye on this one—Asimov may be a small step toward AI that really understands software, not just writes it.
