The HMS Blog

AI coding agents work reliably only if they can navigate the codebase and adhere to established development processes. Christoph Bergen and Robert Bauer explain the role that architecture, documentation, automated tests, and an advanced agent harness play in this process.

AI coding agents don't become reliable simply by using better models. Using the Superpowers workflow as an example, this article shows how context, testing, reviews, and clear decision-making processes contribute to reproducible results.

How Do AI Coding Agents Become Reliable? This article explains how Loop Engineering uses feedback, testing, and clear termination conditions to turn one-off model responses into controlled, adaptive workflows.

A practical guide to reducing context waste, controlling token usage, and improving the quality of agent-assisted software development.

Free-threaded Python enables true parallel thread execution, thereby changing the requirements for modern Python applications. This article explains where race conditions can arise and how to develop thread-safe Python code using locks, queues, events, and other synchronization mechanisms.

SPN Connect and the World Tour 2025 demonstrated that industries such as pharmaceuticals, media, and logistics are already using AI, AISQL, and Cortex to deliver measurable added value. How can these technologies be used responsibly and in a way that yields measurable results today?

Successfully modernizing legacy systems doesn't start with migration—it starts with transparency. Learn how automated code analysis and AI-powered assessments reduce risks, enable informed decisions, and pave the way for AI-ready software.

AI coding agents are currently transforming the way software is developed. They plan tasks, write and modify code, run tests, and respond to feedback. In practice, however, it quickly becomes clear that a powerful model alone does not make for a reliable agent. For an agent to operate securely, transparently, and reliably, it needs the right environment. This environment is called a harness. The task of building, operating, and continuously improving it is known as harness engineering.

Migration sounds like moving. But that is precisely the problem today. If you simply translate legacy code 1:1 into a new language, you carry over architectural problems from the past. The result: modern code built on a foundation that wasn't designed for AI agents.