@AYi_AInotes: In my early years, I remember a type of person silently admired in the codebase — they could spot N+1 through ten layers of call stack, and point out in a flame graph which function was called three extra times. Today, the Codex Skill that Greg Brockman retweeted makes this no longer a privilege of the few…

In my early years in the industry, I remember there was a type of person silently admired in the codebase — they could spot N+1 at a glance through ten layers of call stack, and point out in a flame graph which function was called three extra times. Today, the Codex Skill that Greg Brockman retweeted, makes this no longer a privilege of the few. Why has performance optimization been so scarce in the past? You need to know how to pull flame graphs with Chrome DevTools, run profiles with Node --prof, read perf reports. You need an almost instinctive sensitivity to asymptotic complexity, able to recognize what O(n²) looks like in deeply nested code. You also need to have experienced hundreds of real production incidents, knowing which patterns blow up under millions of data points. These three abilities combined are something that only a decade of project experience can reliably produce — they are scarce resources in a team, and the salary premium comes from there. The Complexity Optimizer that Greg shared is a Codex Skill built by a community developer. One command: npx --yes codex-complexity-optimizer to install, then in the project root tell Codex "analyze my codebase", and it runs in seconds. It specifically digs out hidden pitfalls like O(n²), O(n*m), N+1, loops within loops, scanning entire tables on every render — each finding is precise with file + line number + current complexity + optimized complexity + recommended fix + risk level. The most important design choice: by default it only reports and does not modify code. Each finding is marked low or medium risk, and it tells you what tests to add before going live. In other words, AI does not bypass human decision-making — it prepares all the information needed for humans to make decisions. But what really interests me about this Skill is not what it can do, but what it signifies. Over the past two years, the story of AI writing code has focused on making code writing faster. But the speed of writing code has never been the real bottleneck for developers. The real bottleneck has always been seeing the problems you can't see — architecture risks, performance pitfalls, security vulnerabilities, dependency traps. These things rely heavily on personal experience accumulation and are concentrated in a few senior developers. The real signal of Complexity Optimizer is that this type of capability, which previously needed ten years of experience to output, has for the first time been compressed into a Skill that can be invoked with a single command. Once this path is established, the next wave of Skills won't be far behind — security audit, dependency risk scanning, architecture decay detection, memory leak reconnaissance — all will flood in. In summary, the moat of senior developers won't disappear, but its definition is changing. The old moat was the ability to see problems. The future moat is the ability to judge whether AI's proposed solutions can be implemented in your business scenario. Ten years of experience is being compressed into a single npx command. This might be starting today.