AI model limitations in long-term code maintenance

HumanLayer's GitHub project discusses how AI coding models cannot maintain codebase quality long-term without significant human oversight and steering. The advanced context engineering approach aims to address fundamental limitations in using AI for sustained code maintenance.

"A method that scales beautifully in a forty-minute YouTube demo on a greenfield app is not a method that scales across a real enterprise codebase with twelve teams, a decade of decisions baked in, an

Software Factories Fail Despite AI Agent Acceleration

An article examines why software factories relying on AI coding agents fail in practice, arguing that while these tools accelerate development, they degrade codebase quality over time. The piece contends that "lights-off" factories lack the model sophistication needed to maintain proper code standards.

It started with a core frustration: how do we get better quality, deeper awareness, and precise structural information out of our large codebases? That research led me to build a rich code knowledge g

So I get a 31-file PR opened for review and I realise that I don't really care that much about it as it's no longer my thing. The codebase that was once almost intimate becomes foreign. Alien even may

The # Debian # Linux resolution on # LLM use is nonsensical. It says copyright of # GenAI material is unknown, but that individual contributors must justify their licensing decision. Clearly nonsense

The # Debian # Linux resolution on # LLM use is nonsensical. It says copyright of # GenAI material is unknown, but that individual contributors must justify their licensing decision.

If you have ever wanted me to answer some of your questions at work or help you figure something out... You can! I still have a small consultancy on the side, and I will happily bill you for just a co

Software should work, and talking about it needn't be boring