The best feature of @xai Grok Build right now is how it handles subagents and personas.
Most people still treat the model like one very smart intern that has to do everything at once. Grok Build took a different route: it gives you actual typed workers with defined roles and real restrictions.
There are three main subagent types you can spawn:
• general-purpose → full capabilities. Can read, write, run commands, and spawn other agents.
• explore → deliberately read-only. Fast and safe when you just need to dig through a codebase without risk.
• plan → also read-only, but optimized for high-reasoning architectural thinking. It comes back with structured plans and points out critical files.
The fact that two of the three are read-only by design is intentional. You don’t want your researcher or your architect accidentally modifying code while it’s still exploring or thinking.
On top of the types, you have personas:
Personas like implementer, reviewer, security-auditor, or design-doc-writer carry actual contracts. The implementer knows it’s supposed to receive a review file and produce a summary file. The reviewer knows the exact structure it has to use for bugs, suggestions, and nits.
This changes the game. Instead of one model trying (and usually failing) to be good at research, implementation, review, and security at the same time, you get real specialization and isolation.
The primitive is still early, but it already feels meaningfully different from most “multi-agent” setups I’ve tried.
I’ll post the second thread about what /implement actually does with these pieces.
