30 posts found
Agents are multi-step workflows. Logs aren't enough. You need traces to understand causality, evaluate quality, and debug failures. This article shows how to instrument agents with OpenTelemetry and build lightweight eval pipelines from traces.
Model Context Protocol (MCP) connects agents to tools, but the security story is still immature. This article shows how to build a production tool gateway that enforces least-privilege permissions, input validation, rate limiting, sandboxing, and prompt-injection defenses.
Teams are moving from one big agent to many small agents. They need shared tool access and a clean way to talk to each other. Here's how MCP and A2A fit together without creating a mess.
Once you have many agents delegating work, debugging becomes the main cost. You need to answer basic questions fast: Which agent made the call? Which tools did it touch? What did it see? What did it cost? This is now a common pain point in 'agent mesh' style systems.
How to catch agent bugs before they hit production by treating tools, prompts, and workflows like code.
A practical guide to rolling out new agent versions without breaking production, using patterns from DevOps and feature flags.
A hands-on guide to designing CI/CD pipelines that treat agentic AI (multi-step, tool-using agents) as first-class deployable units, not magical black boxes.
Show, in concrete terms, how to let AI agents touch real infrastructure without turning your cluster into a sandbox accident.
How to build agents that fail safely instead of failing loudly.
How to stop your agents from doing unsafe or surprising things when they call tools like email, payments, or internal APIs.