Cortex 2.7
🚀 Feature rich AI/ML release! Cortex 2.7 updates 6 Enterprise AI MCP Agents with huge performance gains, improved chat session management, updated context handling, and updated authentication for 2 MCP backend connectors.
VSCode and Backend
For the VSCode and backend components, we added support for enhanced session management so users can have more control over their chats and responses. 💬 Updated Atlassian MCP connections to make it cross functional and portable across platforms.
This release also includes improved context tracking across the entire repository. 📂 We include cross-file context in performing security scans and while generating code. This largely expands Cortex’s role as a full-fledged coding agent and brings it on par with other well-established coding agents.
AI Enhancements
We achieved substantial accuracy improvements across the production MCP agents, with the most significant gains reaching +61 percentage points for the Slack agent and +43 percentage points for the Google Workspace agent. 📈
The Slack agent demonstrated the most dramatic improvement, jumping from near-zero accuracy (2.34%) to 64.06%, a 2,637% relative increase. 🔥 Similarly, Google Workspace and GitHub agents showed substantial gains exceeding 200% relative improvement.
As seen in the chart above, we show the number of Agent capabilities plotted with improvement percentage. 📊 This varied significantly across the six supported MCP agents. Notably, the MS365 agent, with 44 capabilities, was the only agent to show negative performance change (-1.04 percentage points).
This result may indicate that larger capabilities introduce optimization challenges: increased prompt complexity and greater potential for confusion. 🧠 Also, the Slack agent’s exceptional improvement despite having fewer capabilities suggests that optimization effectiveness may be inversely related to capabilities. However, the Azure DevOps agent and Google Workspace agent present counterexamples, indicating that capabilities alone do not determine optimization success.
We are delving deeper into these complex problem areas and we are only getting started. More innovations to come!
Pervaziv AI
#aiml #innovations #mcp #agents #codingagents #authentication #PervazivAI
