when software flows like water, its boundaries will soften and cease to exist. 'formless when off, bespoke when on' will just be what computers definitionally do in the future. the period with human authored constructs for machines to operate in will be the historical outlier. this was easier to conclude in ~2022 and then outright resisted in the now waning centaur era because it's a simplistic 'line-goes-up' take – and ofc the median counter to this was that things like repeatability, guardrails, security, etc are that properties that dynamically generated software cannot have at any meaningful cost efficiency. with the progress we're seeing now in the 'output binary' mode of operation (constructing svgs -> 'coding' JS videos better than diffusion), it seems obvious that it can create whatever is optimal in an economically viable fashion; if familiarity and idempotency is optimal then that's what it would do. if it's optimal to cache parts of the software then it will do that too. it’ll all be very amorphous.