All Posts

Some companies have started grading developers on how many AI tokens they consume. It's the latest in a long line of well-meaning productivity metrics that reward exactly the wrong behavior—and this one quietly degrades code quality, burns money, and teaches your most capable engineers to sabotage the very thing leadership is trying to accelerate.

I keep arguing that AI is going to shrink IT departments hard — fewer coders, fewer business analysts, fewer scrum masters, fewer QA testers. That part is real, and I won't dress it up. But there's one genuinely new role emerging in the rubble: the person who takes a freshly AI-built feature, plays with it, and imagines the version that's actually better. There will be nowhere near enough of these seats to replace what's lost, the skillset barely overlaps with the jobs disappearing, and the responsibility is enormous.

The Model Context Protocol promised universal LLM tool integration. Then developers hit the context wall. Here's what went wrong and what's replacing it.

Large context windows are (little white) lies. Here's how to architect your code, structure your prompts, choose your tools, and even pick your programming language to maximize what you can accomplish within the brutal reality of effective token limits.

With AI writing over 90% of boilerplate code, the software developer's job has fundamentally changed. This article explores why the remaining 10%—architecture, context management, and organization—is the most difficult, high-leverage work that AI can't yet touch.