We're excited to announce two powerful new capabilities coming soon to Box Automate that give workflow authors greater control over how AI agents reason and perform within their workflows: Model Selection in Automate and per-outcome Pro Mode Override.
Model Selection in Automate
Today, Box Agents in Automate use a platform-selected AI model for all workflow executions. With Model Selection in Automate, workflow authors can now choose the AI model the Box Agent uses that best fits their specific use case, directly within the workflow builder.
Different models excel at different tasks. Some are optimized for speed and efficiency on structured extraction; others are better suited for nuanced reasoning and complex analysis. Giving workflow authors control over model selection means they can tune agent performance to match the demands of each workflow, without compromise.
With Model Selection in Automate, workflow authors can:
- Choose from AI models enabled by your Admin when configuring a Box Agent step in a workflow
- Match model capabilities to the specific requirements of each workflow use case
- Align model selection with the same controls available in AI Home and other Box surfaces
Pro Mode Control in Automate
Box Automate is also introducing a per-outcome Pro Mode toggle for the Box Agent within workflows.
Today, Pro Mode, which enables deeper reasoning and more capable models, is on by default when using the Box Agent. With the new Pro Mode Control, workflow builders can now toggle Pro Mode on or off for each individual outcome step within an Automate workflow. So the Box Agent can run in standard mode for a simple extraction step and switch to Pro Mode for a complex analysis step, giving you more precise control over AI usage within a workflow.
With Pro Mode Control, workflow builders can:
- Enable or disable Pro Mode on a per-outcome basis within a workflow when using the Box Agent.
- Run cost-efficient standard mode on straightforward extraction or classification steps
- Activate Pro Mode only where deeper reasoning is needed (e.g. complex analysis or multi-document synthesis)
- Build more sophisticated, cost-optimized multi-step agentic workflows
Why It Matters
Enterprise customers building multi-step agentic workflows need precision. A single workflow might include a simple document classification step followed by a complex cross-document analysis, and those two steps have very different performance and cost profiles. Until now, there was no way to tune them independently.
Together, Model Selection and per-outcome Pro Mode Control give workflow authors the granular control they need to build workflows that are both powerful and efficient — optimizing agent performance at every step, not just at the agent level.
Like all AI features, Box AI can make mistakes. Box Automate is designed to help you build human oversight and review steps directly into your workflows, so your team stays in control of every outcome.
Availability
Model Selection and Pro Mode Controls in Box Automate are coming soon. These capabilities will be available to Enterprise Advanced customers with access to Box Automate's full suite of agentic workflow automation features. Any agent run within an Automate workflow consumes Box AI Units.
Resources
- Join the community: AI Content Workflows User Group, and connect with others building content workflows with agentic AI
- Box Automate documentation