3 best practices for building software in the era of LLMs

What this post added

This post introduces three best practices for building software in the era of LLMs: 'Never trust, always verify' (treating AI-generated code with scrutiny, similar to junior developer code, and using tools like GitLab Duo Code Review to enhance human judgment), 'Prompt for secure patterns' (emphasizing the importance of detailed, security-aware prompts, referencing OWASP best practices, and training developers in prompt engineering), and 'Scan everything, no exceptions' (advocating for continuous, integrated automated scanning like SAST, SCA, and Secret Detection to mitigate risks from increased code volume and AI generation). It positions GitLab's platform and AI features as integral to embedding security within the development workflow to balance speed and security.