


Dead Code seeks to extract the good ideas from the chaos of modern software development. Hosted by @jardonamron.


Dave Rooney suspects small user stories worked because they kept a developer's cognitive load manageable. An LLM can hold far more context than one person can. Typing is cheap now too, so tests can go in as a safety net from the very start.


People who refuse to pair with another human are happy to pair with an LLM, and Dave Rooney finds that funny. He was pair programming before he’d ever heard of XP. These days his partner is Claude, which asks him questions and writes tests first that describe expected behavior.


"There's no substitute for actually understanding the problem that's at hand." Varun Gandhi says engineers carry old rules of thumb, like assuming the network is the slow part. On AWS, the network may be plenty fast, and the real time sink is processing or heap allocations.


A ticket that would have been eight small stories took Dave Rooney about two hours with Claude. It cleared PR review and reached the testers in the UK within roughly 24 hours. He figures the usual story-splitting approach would have taken two full business weeks.


An LLM makes it easy to optimize a big chunk of a codebase in one go. Varun Gandhi describes what that does to the rest of the team. A 10-PR stack with thousands of lines per commit is like a cache invalidation for everything your coworkers knew about the architecture.


When an agent runs your 30-second git push in the background, nobody sits there waiting on it. Varun Gandhi says that makes slow tools harder to fix. Fewer people complain, so engineers lose the leverage they need to argue for making things faster.


Race conditions used to scare engineers into getting concurrent code right the first time. Varun Gandhi worries that LLMs change that math. If a model can debug a race condition later, teams may get comfortable shipping more bugs now.


LLMs have left software with fewer excuses to be slow, and Varun Gandhi agrees. He still expects to run into a lot more slow software for a while. Shipping suboptimal code has never been easier, and he thinks the pendulum will take time to swing back.