resources
A thought experiment that reveals more than most audits: if you had to write an AGENTS.md thorough enough to onboard a new senior engineer without a single tribal-knowledge conversation, what could you write — and what would you realize you don't actually know?
Engineering has moved fast with agentic tools. Legal and compliance haven't caught up. Here's what's in your AI pipeline that your risk team needs to know about — and how a well-designed context layer turns a compliance liability into a compliance asset.
Principles are useful. Recipes are actionable. This is an opinionated, task-by-task breakdown of how to allocate your context budget across six common SDLC agent tasks — with specific token targets and what happens when you exceed them.
AI-assisted development is producing a new generation of legacy codebases in record time. Vibe-coded repos with no AGENTS.md, no architecture doc, no decision registry. Six months in, the original session context is gone and nobody knows why anything works.
AI-assisted code review finds syntax errors, style violations, and obvious bugs at machine speed. The things that actually cause incidents — intent drift, constraint violations, architectural boundary crossings — sail right through. Here's the anatomy of the miss.
Context sizing tells you how to pack a window. But what happens when the window closes? The amnesia between agentic sessions is the most expensive hidden cost in AI-assisted software development — and almost nobody is engineering around it.
On every high-performing AI-native team, someone has quietly become the person who makes the AI actually work. This article names that role, defines its responsibilities, and argues it deserves a title — not just tribal status.
How AI agents that gather, structure, and validate requirements don't just accelerate delivery — they fundamentally close the loop between what stakeholders want and what developers ship.
There's a quiet frustration spreading through engineering teams that have rushed to adopt AI coding assistants. The demos looked effortless: a clean prompt, a few seconds, and a perfectly working feature drops into place. But when engineers try to apply the same tools to a codebase that's been in production for seven years — one that predates Docker, runs on a framework EOL'd in 2019, and has ten layers of undocumented workarounds — the magic evaporates.
Larger contexts don't make models smarter — they make them more confidently wrong. Here's the physics of attention decay, why your AI pipeline is hemorrhaging signal, and how to engineer around it.