Key takeaways from the Dreamforce 2025 developer keynote: Agentforce Vibes for agentic development, Model Context Protocol (MCP) for predictable agents, Apex test and tooling improvements, LWC local preview, and Scale Centre for org observability.
Overview
Dreamforce 2025 emphasized Salesforce’s continued investment in AI-driven developer experiences. The developer keynote focused on three major themes: agentic development with Agentforce Vibes, predictable agents powered by the Model Context Protocol (MCP), and a shift toward conversational UX. Several platform improvements for Apex, LWC local development, and observability were also announced.
Agentic development with Agentforce Vibes
Agentforce Vibes extends Agentforce for developers by enabling true multistep, agentic reasoning. Developers can provide high-level, natural-language instructions and Vibes will generate operational code, create related test classes, and even iterate to fix errors.
- Integrated into VS Code via the Agentforce extension for a seamless developer experience.
- Generates Apex classes and tests automatically from prompts.
- Performs multistep workflows: requirement analysis, code generation, test creation, debugging, and making code deployment-ready.
Predictable agents and Model Context Protocol (MCP)
Salesforce introduced the Model Context Protocol (MCP) to enable agents to discover and orchestrate the tools they need. MCP runs locally on its own server and helps agents access org metadata, runtime insights, and external services. A pilot was announced for Spring ’26.
Conversational UX
The keynote demonstrated chat-driven, conversational interfaces where agents guide users through complex tasks. Grounded in business-specific knowledge, these agents produce accurate, personalized, and actionable responses — useful for campaign management, troubleshooting, or code analysis.
Apex: runRelevantTests and other enhancements
Salesforce announced runRelevantTests, which optimizes deployments by running only tests relevant to changed code. The test discovery API and the Cursor class are now GA, and fetchPage improves pagination for smoother results.
Sample syntax shown during the keynote:
@isTest(testfor="ApexClass:ContactService,ApexTrigger:ContactTrigger")
public with sharing class MyDemoTestClass {
// …
}
Scale Centre and Apex Guru
Scale Centre and Apex Guru, previously limited to Unlimited Edition, are now available to all customers. Scale Centre provides performance metrics and highlights hotspots across Apex, Flows, and database usage. Integration with Agentforce enables conversational diagnostics and remediation suggestions. During the demo, agents posted alerts to Slack and used MCP to pull context and surface targeted fixes.
LWC local development and live preview
LWC gained a local preview feature so developers can see and test components directly from VS Code before deploying to an org. This reduces iteration time and lowers the risk of deployment issues.
FAQ
- How many premium AI calls does Vibes allow? Each org gets 50 daily premium model calls (GPT-5) before falling back to Qwen3.
- What is the purpose of MCP? MCP enables agents to automatically determine and call the tools they need to complete tasks, by providing local context and access to org metadata and runtime data.
Key takeaways
- AI agents are moving from single-response helpers to multistep, agentic workflows.
- MCP will allow agents to discover tools and run locally — making agent behavior more predictable and secure.
- RunRelevantTests and test discovery API will speed up deployments and CI/CD workflows.
- Scale Centre and Apex Guru become accessible to more customers, improving org observability.
- LWC local preview reduces development friction and accelerates delivery.
Why this matters
For Salesforce admins, developers, and business users, these announcements signal a shift toward more automated, precise, and conversational workflows. Developers can iterate faster with automated code generation and local previews; admins benefit from better observability and targeted recommendations; business users gain faster, more contextual answers from agents built on org-specific knowledge.
Expected availability: several features are already rolling out while others (MCP pilot, runRelevantTests enhancements) are slated for Spring ’26.