Bad ideas that aren't bad anymore

A software leader argues that AI is rapidly changing the costs behind long-standing engineering rules, so experience-based instincts can become outdated. On his team, engineers now use Claude to produce pull requests, then review the work themselves; CI and Greptile provide additional checks before they merge. This has reduced delays without removing scrutiny, and the ticket owner may be better positioned than a teammate to assess whether the solution fits the problem. The team has also split work across separate streams without the knowledge silos he once feared. AI can quickly brief a colleague on another person’s project, making it easier to join in; the main context to learn is the problem and intended solution, not every detail of the code.

These changes also reshape his career advice. Rather than specializing deeply in a particular technology, engineers should build strong fundamentals in architecture, abstraction, and judgment, while developing skills in design, product work, user research, and business understanding. He is reconsidering other conventional warnings, including rewriting systems, building rather than buying, prototyping several implementations, and adopting unfamiliar tools. His broader point is that these practices were risky or inefficient because of specific costs - such as slow onboarding, scarce expertise, or lengthy development - and AI can sharply reduce those costs. Instead of treating old rules as universal, engineers should ask what made an idea seem bad and whether that cost still applies.