Category
7 articles in this category
Coding agents can write code and run it against real dependencies. A human still has to decide whether the change works. Signadot Plans and Skills close that loop: encode "correct" as a reusable validation plan, then let an agent create a sandbox, run the plan, catch a cross-service regression, fix it, and re-run until it's green before a human sees the PR.
Configure agentgateway for tool allowlists, release service accounts, public and admin access, read-only assistants, federated servers, and ExtMCP guardrails.
An AI agent is a machine that holds your credentials and can spend money. In this post, I'll show you how to enforce a shared token budget per agent with agentgateway, and why a local rate limit stops being a real limit as soon as you add another gateway replica.
Coding agents are quite good at writing code now. Any of the agents can easily add a function, wire up an endpoint, or open a pull request for you. But there's a gap that you run into in practice. They write code that looks perfectly correct in a single-service context, and then it breaks as soon as it touches real dependencies; either another microservice, a database, a message queue. The agent has no way to know, and honestly, neither do you, until something fails downstream.
Preflight is a VSCode extension that reviews your code changes and generates PR summaries using AI — all before your code leaves your machine.
What is an AI agent actually made of? A practical breakdown of agent anatomy — planning, reflection, tool use, memory, and multi-agent patterns — with diagrams and real examples.
In this article, I will demonstrate easy steps of preparing and launching compute-intensive AI applications on OCI Ampere A1 shape to take advantage of the best-in-class AI inference performance of Ampere Altra processor and Ampere Inference Optimizer (AIO).