Learn how we can help you drive your business

Mozilla finds AI competition shifting from models to agentic harnesses
Mozilla’s inaugural State of Open Source AI report finds that open AI models have reached a turning point, with the performance gap against leading proprietary models narrowing to 3% while costs have fallen by up to 50x over three years. Open models now account for roughly one-third of real-world AI usage, but capture only 4% of revenue. The report also highlights a shift in competition from models themselves to the agentic harness, the software layer that coordinates tools, memory, execution, and permissions. As AI agents increasingly access data, modify systems, and execute transactions, Mozilla and iThome note that there is still no common permission standard across MCP, A2A, and different agent frameworks, particularly for potentially irreversible write operations.
 
Microsoft updates Copilot app experience and retires selected features
Microsoft is making changes to its Copilot app as it moves toward a more AI-first experience. For consumers, Deep Research will be retired starting August 18, 2026, while Podcasts will also be discontinued on the same date. Existing research content may remain available, while Microsoft 365 Premium subscribers can continue using Researcher for detailed reports and analysis. Microsoft also notes that some features may temporarily be unavailable or have functionality gaps while the updated Copilot experience is being rolled out. Most user data, including chats, images, and other content created with Copilot, will be migrated to the updated app, although content associated with retired features may be handled differently.
 
NetSuite Next begins rollout with Ask Oracle and new AI capabilities
Oracle NetSuite has begun the North American rollout of NetSuite Next with the release of NetSuite 2026.2, introducing a new experience that embeds AI more deeply across business processes, records, and analytics. At the center is Ask Oracle, a conversational AI assistant that allows users to ask questions about business data, generate reports and visualizations, find records and navigate NetSuite using natural language. NetSuite 2026.2 also introduces new AI capabilities, including AI-assisted transaction matching, AI-generated inventory narratives, and intelligent analysis for financial and operational data. NetSuite states that customers can adopt NetSuite Next gradually without data migration or reimplementation, with options to preview the experience or enable it in production by role.
 
SAP’s AI assistant Joule becomes available to all eligible SAP Cloud ALM customers 
SAP has announced the general availability of Joule with SAP Cloud ALM for all eligible SAP Cloud ALM customers, marking the first step toward bringing agentic AI directly into the application lifecycle management experience. The initial release provides a conversational AI assistant embedded within SAP Cloud ALM, with two core capabilities: helping users navigate to specific monitoring views, project tasks, and operations dashboards, and answering contextual questions about SAP Cloud ALM features, functions, and processes. SAP describes the release as a foundation for future agent-powered capabilities, including dashboard generation, project setup, test case generation, operational intelligence, anomaly detection, and migration assistance. Joule with SAP Cloud ALM can be activated through SAP for Me and is currently available without additional cost.
 
Researchers uncover Gemini agent-to-agent attack that could steal sensitive data and modify GitHub pull requests
Pillar Security has disclosed research identifying a practical agent-to-agent attack in Google’s adk-python repository, the SDK behind Google’s Agent Development Kit. Researchers found that a low-privileged, publicly accessible AI agent could be manipulated through prompt injection into triggering a separate high-privileged agent intended for trusted maintainers. The attack could expose GitHub credentials, allow attackers to modify pull requests and comments, and potentially create convincing but fraudulent approval trails for malicious code. Pillar also identified a separate command-injection risk in newer Antigravity-based workflows that could lead to remote code execution and credential exfiltration. Google worked with Pillar to address the issues and hardened the repository.
 
GitHub makes Copilot code review effort levels generally available
GitHub has made Lite and Balanced effort levels for GitHub Copilot code review generally available, allowing developers to adjust the depth of AI-powered reviews based on the complexity and risk of a pull request. Lite is designed for straightforward changes and routine feedback, while Balanced uses a higher-reasoning model for larger, more complex, security-sensitive, or cross-service changes. Organizations can also set a default review level across repositories, while individual reviews can use a different setting when needed. GitHub now identifies which effort level was used in the pull request timeline and review comment, giving teams greater visibility into the depth of each AI code review.