Google PM open-sources Always On Memory Agent, ditching vector databases for LLM-driven persistent memory. Google open-sources an always-on LLM memory agent that stores structured memories without vector DBs, simplifying persistent agent memory for production use. This removes a major infra component for persistent sessions and changes how you model context for agents in production (Principles 06 & 11).

Cursor launches Automations to trigger agents from code changes, Slack, or timers. Cursor rolls out Automations to fire agents from code commits, Slack messages, or scheduled timers to automate developer workflows. That makes agentic workflows event-driven and CI-integrated, turning agents into reliable hooks in delivery pipelines rather than ad-hoc helpers (Principle 09).

Visual Studio Code previews agent plugins. VS Code 1.110 previews agent plugins, browser tools, and a real-time agent debug panel to give developers control and persistent session memory. Editor-native agent visibility and debugging turns agents into first-class developer tools and strengthens human checkpoints during delivery (Principles 03 & 15).

Codex Security: now in research preview. OpenAI debuts Codex Security, an agent that grounds vulnerability discovery in project-specific context, validates findings, and proposes safer fixes to reduce triage noise. Project-aware security agents demonstrate how outcome-focused tooling can automate detection while raising the bar on actionable, verifiable recommendations (Principles 14 & 06).

Beyond ‘Prompt Thrash’: A Framework for Moving Agents from Demos to Production. The framework prescribes treating agent quality as engineered, using Outcome Specs and Convergence Loops to rapidly evaluate and prune configurations for production readiness. It gives a practical playbook for moving agents from experiments to dependable infrastructure by codifying specs, evaluation loops, and acceptance criteria (Principles 06, 14 & 16).