Drawing on insights from more than 12,000 previous Agentforce implementations, the platform introduces Agent Script, a human-readable scripting language that lets teams define complex agent behavior with precision, including conditional logic, guided controls, and workflow hand-offs. Integrated with the configurable Atlas Reasoning Engine, Agent Script balances the creativity of large language models with structured business logic and now supports multiple models, including Google Gemini, OpenAI, and Anthropic.
The platform also features Agentforce Voice, which delivers ultra-realistic, low-latency, voice-first interactions. Agents can execute actions such as updating CRM records, triggering workflows, or calling APIs, while maintaining live transcription for seamless human agent takeovers. Voice capabilities integrate with Salesforce and leading contact center platforms including Amazon Connect, Five9, NiCE, and Vonage.
Agentforce Builder allows teams to design, test, and refine agents in a single workspace, supporting multiple views including doc-like editors, low-code canvas, and script view. Its AI coding partner, Agentforce Vibes, accelerates development by collaborating contextually with project data. Meanwhile, Intelligent Context, powered by Data 360, transforms unstructured content including PDFs, tables, images, and flowcharts into actionable data, enabling agents to provide accurate, business-specific responses quickly.
John Thompson, VP of Sales Strategy and Operations of Reddit said, “Since implementing Agentforce, Reddit has been able to deflect 46% of customer support cases and cut the average resolution time from 8.9 minutes to just 1.4 minutes, which is an 84% improvement. This efficiency has allowed us to provide on-demand help for complex tasks, boost advertiser satisfaction scores by 20%, and free up our human reps from repetitive questions.”
