For years, much of the software industry worked on making developers more productive. Better languages, frameworks, libraries, cloud services and development tools progressively reduced the work required to turn an idea into code. AI has accelerated that process dramatically. What once took weeks can sometimes take days or hours.
But software production is a chain, and making one part of that chain dramatically faster does not necessarily make the whole system faster. Code still has to become a running application. It has to be deployed, connected to infrastructure, observed, updated and recovered. As the time spent producing code falls, the relative weight of everything surrounding it increases.
This is why some of today’s productivity gains can feel incomplete. We can watch an agent produce an application in minutes and then find ourselves back in a familiar world of environments, permissions, configuration and deployment. The contrast makes these steps feel more cumbersome than they did before, even when nothing about them has actually become worse.
Every major increase in productivity eventually reorganizes the work around it. AI-assisted coding will be no different.
The interesting question is no longer only how much faster we can produce software, but what has to change when producing the code is no longer the scarce part of the system.