Salesforce and Anthropic launched a Claude Cowork plugin that lets sellers query and update live CRM data through 37 pre-built sales skills without opening Salesforce. Arga Labs raised 10 million dollars to build digital twins of enterprise software so agents can train on realistic environments. AWS released Amazon Bedrock AgentCore Evaluations, a service that reads OpenTelemetry traces from any agent framework and applies built-in quality checks without per-SDK setup.
Salesforce and Anthropic launched a Claude Cowork plugin called Salesforce in Claude built on 37 pre-built sales skills, that allows sellers to query, update and act on live CRM data without opening Salesforce. The plugin builds on Salesforce's earlier API and MCP server work by centralizing authentication and permissions. It is currently live for select pilot customers, with an open beta planned for September.[1]
Arga Labs raised 10 million dollars in a seed round led by General Catalyst to build practice copies of business software like Salesforce and Workday. The startup creates digital twins of enterprise applications with their permission systems and webhooks intact, allowing agents to be trained in a realistic environment rather than against stateless API endpoints. The approach aims to address the reinforcement gap by letting the environment be modified and replayed as often as training requires.[2]
AWS introduced Amazon Bedrock AgentCore Evaluations, a framework-agnostic service that lets teams grade any AI agent uniformly. The service reads OpenTelemetry traces to reconstruct an agent's session—identifying requests, model calls, and tool invocations—and applies built-in checks like goal-success rate and correctness. It works out of the box with frameworks including LangGraph, LlamaIndex, the OpenAI Agents SDK and the Claude Agent SDK without requiring custom setup.[3]
