The AI Productivity Gap

The strongest thread running through the comments is that AI has shifted the bottleneck from writing code to absorbing and controlling it. People described spending more time steering agents, reviewing swollen pull requests, and debugging failures that look polished on the surface but break in strange ways underneath. Several said the output feels deceptively trustworthy because it mimics the visual cues of competent code while hiding weak reasoning and unnecessary abstraction. That creates a nasty accounting problem. Individual developers feel faster because text appears quickly, while teammates inherit more review load, more maintenance risk, and more junk work from speculative fixes and over-refactoring. There was still a clear pro-AI position, but it was narrower and more operational than the usual hype. The credible positive reports came from teams with tight harnesses, strong specs, automated checks, adversarial review, and engineers experienced enough to keep models inside well-lit boundaries. In those setups, commenters reported around 2x gains, better documentation, and in some cases fewer bugs. The common theme was that AI does not remove the need for engineering discipline. It raises the premium on it. People also noted that gains vary a lot by role and company size. Juniors and new team members may benefit most as learners and codebase navigators, while senior engineers still spend their time on judgment, tradeoffs, and system ownership. Smaller teams may see bigger end-to-end gains because they have fewer organizational chokepoints. Overall, the comments settled on a blunt view: coding got faster, but unless your process is built to constrain and verify machine-generated work, the saved minutes reappear as supervision, review, and downstream cleanup.