← Latest Update

Agent Ops: Hardening RAG, Debugging Browsers, and Building Agentic Workflows

We Ran Real Attacks Against Our RAG Pipeline. Here’s What Actually Stopped Them. runs real-world attacks against a ChromaDB+LM Studio RAG stack and shows three defenses—embedding anomaly detection, access-controlled retrieval, and prompt hardening—that stop the exploits. Outcome engineers gain a concise mitigation playbook to harden retrieval pipelines and monitor embedding drift, directly addressing the system-level immune controls teams must build (Principles 14 & 15).

Let your Coding Agent debug the browser session with Chrome DevTools MCP launches Chrome DevTools MCP support so coding agents can attach to live browser sessions, execute DevTools commands, and debug with user consent. This enables agents to perform observable, actionable steps in the browser—design patterns you can adopt to give agents controlled execution capabilities and verifiable artifacts (Principles 03 & 15).

What is agentic engineering? Simon Willison lays out agentic engineering patterns that shift engineers from single-player coding to goal-driven agents, emphasizing goal specification, tooling, and rigorous verification loops. Treat this as a practical framework for reorganizing teams and workflows around agents and outcome validation (Principles 01, 03, 16).

How to build an AI agent that actually works recommends embedding agents into workflows, inserting agentic loops only where human judgment is required, and assembling deterministic context before model calls. Use its workflow-first guidance to reduce brittle autonomy—design deterministic context layers and clear handoffs so agents deliver repeatable, auditable outcomes (Principles 06 & 02).

Italian startup Alomana raises €4M for its AI operating layer for enterprise workflows announces funding to scale Alo, an AI operating layer that runs end-to-end enterprise workflows across apps and data. This signals more production-grade orchestration layers entering the market—evaluate these platforms as potential islands for shipping repeatable, monitored agentic processes inside your org (Principles 04 & 09).