source: simon willison: quoting andrej karpathy
level: technical
andrej karpathy shared thoughts on how software creation is shifting. he said working software increasingly comes out on a tap. this change makes him feel his own demand for software growing a lot. he can ask for anything now. examples include explainers, visualizers, dashboards, and single-use apps. he mentioned a hyper-specific wandb just for one project. he can also expand test suites, auto-optimize code, and run large research projects with custom html results.
karpathy referenced the jevons paradox. this economic idea says that as a resource becomes more efficient to use, consumption of it goes up instead of down. here, the resource is software development. as ai makes it cheaper and faster to build software, people will want more of it. karpathy sees his own behavior matching this pattern. the ease of generating tools on demand fuels a cycle of wanting even more tools.
the quote was collected by simon willison and posted on his weblog. it highlights a practical shift for developers and researchers. instead of waiting for general-purpose tools, they can spin up custom solutions instantly. this could change how people approach problems. they might tackle more ambitious projects or automate more tasks. the barrier to creating software drops, so the appetite for it rises.
why it matters: ai that generates software on demand could make custom tool creation so cheap that data scientists and developers start building far more bespoke solutions, changing how they work.