Salesforce on October 27 detailed the next phase of its Agent Fabric platform, comprising Agent Evals (beta in November, general availability in January 2027), the API Evals tool, a custom agent scanner beta, and an evaluation-triggered agent kill switch planned for January 2027.
Salesforce announced the acquisition of customer engagement platform Braze for approximately $5.66 billion, or $47.50 per share in cash, a premium of about 58.1%. The deal aims to strengthen Salesforce's AI business as AI adoption reshapes traditional software demand. The transaction is expected to close in fiscal 2028 and does not affect fiscal 2027 guidance. Braze serves about 2,770 customers, and its AI tools help optimize marketing content and channel delivery.
On October 12, Salesforce announced it had signed a definitive agreement to acquire Loops, an AI-powered analytics platform. Loops uses machine learning to monitor business performance, explain metric changes and identify improvement opportunities. The deal is expected to close in early November of Salesforce's fiscal year 2027, subject to customary closing conditions.
At Dreamforce, Salesforce launched Claudeforce with Anthropic, enabling customers to call Salesforce directly in Claude to complete tasks.
Salesforce AI Research and Agentforce released DarwinX, a new framework that uses an evolutionary approach to optimize an AI agent's prompts, tools, skills, and workflows (the harness), raising browser-task completion from 43.5% to 93% without modifying model weights, and improving performance across four benchmarks.
Anthropic CEO Dario Amodei and Salesforce CEO Marc Benioff said enterprise AI adoption is still in its early stages, with only about 5% to 10% of AI's economic value currently being utilized. Benioff said Salesforce has around 7,000 employees using enterprise tools integrated with Anthropic's Claude models and needs to help customers accelerate AI transformation. Nvidia CEO Jensen Huang said market forces will constrain AI safety, while OpenAI CEO Sam Altman expressed confidence that the industry can safely advance technology.
At Salesforce Dreamforce in San Francisco, Nvidia CEO Jensen Huang said AI safety does not require new legislation, as market mechanisms are sufficient to regulate AI product safety and companies should set their own release pace. He called the idea that innovation and safety are opposed a false dichotomy.
Salesforce launched Koa, its first reasoning model, at Dreamforce 2026. Built on Nvidia's open-weight Nemotron model and jointly post-trained by the two companies, Koa focuses on sales, marketing and customer-service tasks, offering an open-weight alternative to closed models in the Agentforce platform to reduce AI spending and safeguard data.