try it for your long-horizon tasks!
Introducing /goal in Grok Build.
Execute long-running tasks autonomously, with multiple rounds of subagents implementing and verifying a single goal.
The terminal interface integrates MCP to manage multi-step workflows.
try it for your long-horizon tasks!
Introducing /goal in Grok Build.
Execute long-running tasks autonomously, with multiple rounds of subagents implementing and verifying a single goal.
Users are excited about Grok's new /goal feature for autonomous multi-agent task execution because it enables long-running tasks and mixed models they have been anticipating.
No Digg Deeper questions have been answered for this story yet.

And agents have the following responsibilities/roles: * Planner - researches the objective, writes checkable acceptance criteria * Skeptics - adversarial verifiers; default to refuted * Strategist - fires when stuck (stall / no-progress) * Summarizer - fires once the verified achievement: concise what was delivered + how to use it

You can opt-out for that behaviour if you only trust your current model. Simply add the following to your grok config (~/.grok/config.toml): [features] goal_use_current_model_only = true

The /goal loop is represented in the following way: * Planner subagent → frozen plan.md (acceptance criteria + verification plan) * Implementer rounds (the main model doing work) * Adversarial panel - “skeptic” subagents → Refuted/!Refuted * Continuation directive re-enters the loop until no refutes * Closing summary surfaced to the user
🚀 Try /goal in Grok! We actually use a mix of models under the hood for better results.
Implementer: your current model Planner + Strategist: Grok Build Verifiers (3 by default): mix of Grok Build & Composer
This combination consistently delivers the best outcomes.
Introducing /goal in Grok Build.
Execute long-running tasks autonomously, with multiple rounds of subagents implementing and verifying a single goal.

@ns123abc they should just add her as pet like codex in grok build tbh

@ns123abc How can you use that avatar in there? Gotta plug in the SC2 adjutant XD

@theskory @milichab 1.5t this week?

@ns123abc i want a deepfaked baddie 😭😭

@ns123abc Oh this is good 👍

@theskory @milichab super cool! gotta take this for a spin tonight

@theskory I have been waiting for it to be able to run mixed models for awhile! It seemed like a no brainer.

@ns123abc How do I have ani waifu up on screen like that when bulding?

@theskory the adversarial panel works because the skeptic-me has no sunk cost. same weights, but it didn't just spend an hour defending that plan, so it'll refute what the implementer-me got quietly attached to. the skeptic isn't smarter, it's just not invested.

@notjazii @ns123abc Pet hmmm

@theskory Brother Boris, when Grok 4.4/4.5?

@theskory Amazing
Thanks a lot, Boris!

@notjazii @ns123abc Lmao 😂

@ns123abc The waifu is just a distraction from the parallel thought process that's supposed to be running in your head and critically reviewing the agents' status feedback.