NinjaOne launched Katana AI (early access), an in-console agentic technician that summarizes device health and recommends fixes technicians must approve, plus a NinjaOne MCP server so external AI agents can reach NinjaOne data and actions under existing RBAC and OAuth.
Now in beta, Checks lets coding agents run CI mid session in credential free, egress restricted microVMs from warm dependency snapshots, with failure logs fed back into the agent's context. Each passing run emits a signed attestation that merge gates can verify instead of re running the test suite…
Cursor added Gemini 4 Argon with high/medium/low/minimal effort levels; at default medium it scores 52.8% on CursorBench, roughly matching Opus 5.5 Medium. Usage draws from the third party Other Models pool at $2/M input and $10/M output…
Claude Code 2.1.299 adds idle compaction for 200K+ token terminal sessions left idle about 54 minutes before the one hour prompt cache expires, plus CLAUDE CODE DISABLE FILE MENTIONS so @ mentions pass paths for Claude to read instead of file contents. MCP allow rules now reach only the server they name…
DanubeData opened a Streamable HTTP MCP at https://danubedata.ro/mcp that turns nearly every public API op into typed tools (VPS, databases, object storage, Rapids, and more), scoped to the API token you attach—no local install. Add read only=1 or ?toolsets=vps,database to narrow what the agent sees; OAuth only clients like ChatGPT custom connectors can't connect yet…
Google ran Gemini 4 Argon unattended for 11 hours on google/adk python via Antigravity /goal mode; maintainers shipped 2 of 6 proposed changes that cut InMemorySessionService prep 40% in long sessions and cold import google.adk by 21%. Four other proposals were rejected in review for fragile caches, private OTel attrs, or synthetic only gains…
Spotted something we missed? Start a thread.