Salesforce introduced Hunter, an Agentforce-native outbound sales agent that autonomously prospects, enriches and sequences using ZoomInfo and Demandbase data.

• Built-in data access: Hunter surfaces ZoomInfo prospect records and Demandbase account intent signals inside Agentforce without requiring a separate export or import step.
• Expanded Docusign partnership: The launch also extends Salesforce's Docusign integration, keeping contract actions inside the same agent workflow.

Profound launched a conversational agent-builder for marketers that lets teams describe a job in plain language, assign tools, and have the AI Marketer run it on a schedule or trigger with visible reasoning on every run.

• Judgment-aware execution: Agents distinguish between mechanical data collection and decisions that require human review, surfacing only meaningful signal rather than raw output.
• Full audit trail: Every run shows its reasoning chain, and outputs requiring publication or outbound send stay gated until the marketer approves.

Cognism launched an MCP connector that lets Claude and ChatGPT query its GDPR-compliant company and contact database directly from the AI assistant interface.

• No API build required: Reps connect through Claude's Settings then Connectors menu, skipping the traditional API integration path entirely.
• Verified data on demand: Queries return Cognism's phone-verified, GDPR-first records rather than relying on whatever contact information the model was trained on.

AWS released two open-source projects for agent development: Strands Decider 2B, a 2-billion-parameter model built to select between predefined options inside agent workflows, and Strands Box, a sandbox for controlling agent execution.

• Purpose-built for routing decisions: Strands Decider 2B handles tool selection, model routing, classification and guardrails rather than generating long-form text, keeping latency low inside agentic pipelines.
• Strands Box isolation: The accompanying sandbox gives engineering teams a controlled environment to run, inspect and govern agent behavior before production deployment.

GitHub's AI Scan for pull requests reached general availability, extending automated vulnerability detection to languages and frameworks CodeQL does not cover, including PHP, Shell, Terraform HCL, Dockerfiles, JSP and Blazor.

• Automatic activation: AI Scan runs on every qualifying pull request when code scanning is enabled, and findings appear alongside CodeQL results in the same review interface.
• Broader stack coverage: Teams using infrastructure-as-code, PHP backends or Java JSP can now get the same PR-level security signal previously limited to CodeQL-supported languages.

GitHub Copilot in Slack and Microsoft Teams now supports querying across multiple repositories in a single conversation, letting engineers track PRs, issues and activity across a full project without switching contexts.

• Cross-repo queries in chat: A single Slack or Teams thread can now pull state from multiple repos simultaneously, removing the need to run separate queries per repository.
• Model updates included: The release also ships underlying model improvements that improve the quality of code-related answers in both integrations.

HubSpot gave customers new controls over AI usage costs, including monthly spend caps, the ability to convert unused seats into AI credits, and flexible payment options tied to actual consumption.

• Spend caps and credit swaps: Admins can now set a hard ceiling on monthly AI spend and redeploy unused seat licenses as AI credits, aligning cost to actual agent usage rather than contracted headcount.
• Flex payment options: Teams can shift between seat-based and consumption-based billing within the same contract, reducing the risk of committing to a usage tier before agent adoption is known.

Adobe added agentic workflow capabilities to Marketo Engage Smart Campaigns, enabling agents to automatically clean form-fill data, normalize lead fields, detect bot activity and route records before they reach scoring or nurture flows.

• Pre-scoring data remediation: Agents intercept messy or inconsistent form submissions and standardize field values such as job title variants before records enter lead-scoring logic.
• Bot and spam filtering: The agentic layer flags suspicious lead records automatically, removing a manual review step that previously delayed campaign execution.