Cleaning up after AI rockstar developers

Cleaning up after AI rockstar developers
4 minutes by Jesse Skinner

Rockstar developers often leave behind complex, unreadable code that cripples the teams that inherit it. AI coding tools carry the same risk, but at a far greater scale, generating massive amounts of tangled code across many sessions with no consistent vision. Jesse argues that the solution is to stay in control, move deliberately, and simplify wherever possible. Good software is still a craft, and humans must lead it.

Context rot makes coding agents confidently wrong
sponsored by Falconer

You know which context to trust, what's outdated, and what only lives in someone's head. Your coding agent doesn't. Falconer gives you a live Knowledge Health score, showing what's stale or contradictory, and helps clean up the context your agents rely on. Get your free report.

A new era for software testing
3 minutes by Salvatore Sanfilippo

AI agents can now run software quality checks that were previously done by hand, or skipped entirely due to time constraints. You give the agent a list of tasks, like testing distributed systems or checking for speed regressions, and it works through them while focusing on recent code changes. This can catch bugs that standard automated tests miss. Better QA may help offset the lower code quality that often comes with AI-generated software.

Life is too short for a slow terminal
6 minutes by Mijndert Stuij

Mijndert explains how to make a terminal and shell feel faster by reducing startup delays and input lag. He recommends avoiding large shell frameworks, caching completions, lazy-loading tools such as nvm and kubectl, using an asynchronous prompt, and choosing a fast terminal emulator. He also shows how to measure shell performance with profiling tools and argues that small speed improvements can significantly improve daily productivity.

Nobody pushed back: Why engineers stay silent until it's too late
6 minutes by Emirhan Yildirim

Most engineering disasters aren't caused by ignorance. The engineers knew. Nokia, Boeing, TSB, Microsoft all had people who saw problems coming, but staying quiet was safer than speaking up. Silence becomes a habit, then a culture, then a system property. Building teams where people can say "this will go badly" without consequences is a structural problem, not a personal courage problem.

Loop engineering
12 minutes by Addy Osmani

Instead of manually prompting coding agents one turn at a time, you can now design automated systems that do the prompting for you. These loops combine scheduling, parallel workspaces, stored project knowledge, tool connections, and separate reviewer agents to find work, act on it, and check results without you driving each step. Both Claude Code and Codex now ship all five building blocks. The catch is that a faster loop also makes mistakes faster, so staying the engineer who understands and verifies the output still matters.

And the most popular article from the last issue was:

• Fast is better than slow