In this video, I took the brand-new Agentforce Builder (now in open beta) for a spin, because it’s a pretty major shift in how we design, control, and refine AI agents on the Salesforce platform. Historically, Agentforce could feel like a bit of a black box—especially when you were trying to chain actions together reliably. What’s changed here is the development experience: Salesforce is giving both admins and developers clearer tools to orchestrate agent reasoning, mixing deterministic logic with LLM-powered reasoning in a way that’s far more predictable. The headline feature for me is Agent Script: a domain-specific language that lets you use programmatic logic (think if/else, variables, and sequences) to make agent behavior more reliable and less dependent on “whatever the model decides at runtime.” If you’re more visual, the new Agent Canvas presents everything as structured blocks, with inline actions to set variables, prep context, and run required steps before reasoning begins. Add in supercharged AI assistance (generate topics/actions/script from natural language), plus much better testing and debugging (full conversation simulation and step-by-step breakdowns), and it’s clear Salesforce is pushing toward “enterprise-ready” agentic AI—where hallucinations and misclassification simply aren’t acceptable.