Blog
Notes and analysis on AI development.
Vercel Sandbox Drives beta keeps agent workspaces beyond disposable VMs
Vercel Sandbox Drives is in private beta, separating short-lived Firecracker microVMs from persistent AI agent workspaces.
ChatGPT Lockdown Mode turns off agent web access for prompt injection defense
OpenAI expanded Lockdown Mode to all logged-in ChatGPT users, limiting web, agent, and file-download paths that can turn prompt injection into data exfiltration.
Microsoft Work IQ APIs go GA June 16 with ten tools for M365 agents
Microsoft Work IQ APIs reach GA on June 16, bringing Microsoft 365 work context, ten MCP-style tools, and Copilot Credits pricing to agents.
Google says Agentic RAG raises accuracy 34% by checking missing evidence
Google Research introduced Agentic RAG for Gemini Enterprise Agent Platform, with cross-corpus routing and a sufficient-context check before answers are finalized.
Claude security harness shows a 7-step path from AI-found bugs to proof
Anthropic released Claude Code skills and a gVisor-based vulnerability discovery harness focused on verification, triage, and patch validation.
Netskope AI Command Center turns MCP servers into security inventory
Netskope AI Command Center discovers AI apps, agents, MCP servers, local models, and data-store links inside enterprise environments.
Claude’s five-hour error window asks how agents recover from model outages
Claude’s June 2 model error incident shows why AI agent products need retry, checkpoint, fallback, and human handoff design.
Kurrent Capacitor turns coding agent sessions into team memory
Kurrent Capacitor records Claude Code, Codex, and Cursor sessions as shared memory for PR review, recall, and agent evaluation.
Microsoft Scout brings OpenClaw-style autonomous agents into company accounts
Microsoft Scout connects an OpenClaw-style always-on agent to Teams, Outlook, shell, browser, MCP, and Agent 365 controls.
MAI-Code-1-Flash shows Microsoft’s Copilot token-cost strategy
Microsoft MAI-Code-1-Flash is rolling into Copilot with 137B MoE, 5B active parameters, 256K context, and new AI Credits economics.
Salt Code brings MCP policy enforcement into AI coding tools
Salt Code uses MCP to push security policy into Cursor, Codex, Claude Code, and other AI coding assistants before vulnerable code reaches review.
Step Functions adds AgentCore harness calls as approvable agent steps
AWS added an optimized Step Functions integration for AgentCore harness calls, turning agent reasoning into a workflow Task with approval, retry, trace, and token controls.