Now loading...
The latest update for Grok Build now includes a feature that enables the execution of workflows, allowing users to articulate complex tasks in simpler language. Grok takes these descriptions and orchestrates them through multiple parallel agents in the background, providing results once the entire operation is completed. Notably, users enjoy a cost-free session throughout this process.
This new capability is particularly designed for tasks that extend beyond the scope of a single conversation, such as conducting an extensive review of features in a large pull request, addressing numerous issues efficiently, or scanning a codebase for specific bugs. When a task can be broken down into several independent components leading to a consolidated outcome, it’s advisable to use a workflow.
Grok automatically facilitates the creation of a task plan as a concise script detailing the phases involved, the agents tasked for each phase, and how their outputs will be compiled. Each agent is features an clear, focused context. The planning process incorporates checks that a simple run would overlook, such as having independent reviewers validate every conclusion prior to including it in the final summary.
For these operations, Grok assigns a limit of 128 agents, which can extend to 1,024 for more extensive tasks. Progress is continuously saved, allowing users to pause and resume work without needing to redo completed segments. To monitor the activity live, users can execute the command /workflows, which updates them on the progress across different phases and agent performances.
Users no longer need to write the scripts, as Grok constructs the workflows from the initial request, conducts preliminary checks before implementation, and enhances them after each execution. Once a workflow is successful, it can be retained for future use. These workflows are saved in the .grok/workflows/ directory for team sharing, while those in ~/.grok/workflows/ travel with the user across different devices. Each retained workflow is accessible as its own slash command, enabling quick execution, such as /pr-review 5137 for the repeated review of pull requests.
The /deep-research command provides a built-in option, dispatching research inquiries to multiple investigators who verify claims against available sources, culminating in a comprehensive, cited report.
Potential commands to explore include instructions like, “Use a workflow to check each feature in this PR,” “Triage the last 100 Linear issues and provide a prioritized action list,” or “Audit each route handler for absent authentication checks and confirm each finding.”
