Salesforce introduced Hunter, an outbound AI sales agent that pulls Demandbase buying-group and engagement signals directly into prospecting and account-prioritization workflows.

• Demandbase integration brings real-time intent and buying-group data into Salesforce so Hunter can surface and prioritize accounts without a rep leaving the CRM.
• Hunter automates outbound sequences by acting on those signals, reducing the manual work of identifying which accounts are in-market and when to reach out.

Moonshot AI released Kimi K3.1 Pro, a 2.8-trillion-parameter mixture-of-experts open-weight model with a 1-million-token context window, available today in Kimi Code and the Kimi API.

• Kimi K3.1 Pro Highspeed runs at two to three times the speed of the base model with identical capabilities, giving teams a cost-latency tradeoff without switching providers.
• Full weights ship by November 4, making it one of the largest open-weight coding models teams can self-host or fine-tune for agentic SDLC work.

Google has open-sourced the Elephant-Goldfish Model plugin, an internal tool that forces a coding agent to write and defend a design document before touching any code, then routes that document to fresh AI sessions tasked with finding its flaws.

• The plugin separates planning from execution: a design document produced in one AI session is handed to separate sessions that have no memory of the design conversation, reducing confirmation bias in code review.
• Built for agentic velocity risks, the tool directly targets the reliability problems Google says emerge when coding agents ship code faster than humans can review intent.

A new GitHub app in Microsoft Teams lets engineers @mention GitHub to have Copilot investigate a bug notification, write a fix, and open a pull request without leaving the Teams conversation.

• The workflow is three steps: get notified of an issue in a channel or chat, @mention @github to trigger investigation and fix, then review and merge the resulting PR.
• Notifications and Copilot are now unified in a single Teams app, replacing the previous GitHub notifications app that had no agentic action capability.

ZoomInfo launched a native integration with Exa that lets mutual customers bring ZoomInfo's GTM Intelligence into Exa through Exa Connect, and lets developers pull the same B2B data through the GTM.AI API and MCP server.

• GTM.AI is ZoomInfo's headless data layer, connecting its GTM Context Graph to partner platforms and customer-built agents through both a standard API and an MCP server.
• Agents built on Exa can now query ZoomInfo's B2B intelligence at runtime, enabling prospecting and account-research agents to combine live web search with structured GTM data in a single call.

Salesforce introduced Koa, its first CRM reasoning model for Agentforce, built by post-training NVIDIA Nemotron open models on a proprietary synthetic dataset derived from 27 years of Salesforce workflow data.

• Koa is purpose-built for multi-step enterprise CRM tasks such as account planning, case routing and opportunity scoring, where general-purpose models lack the workflow context to reason correctly.
• Built in collaboration with NVIDIA, Koa ships as the default reasoning model inside Agentforce, meaning existing Agentforce deployments get a domain-tuned model without reconfiguration.

GitHub's Copilot usage metrics API now reports model usage from tools including OpenCode, Claude Code, Codex, and Zed that were previously invisible to enterprise usage reports.

• A new `totals_by_client` array breaks out usage by client tool across enterprise, organization and user-level 1-day and 28-day reports, so engineering leaders can see exactly which agents consume Copilot capacity.
• Unrecognized clients are grouped separately rather than dropped, preventing undercounting when new agentic tools appear in a developer's workflow.