@doggintrump this is absurdly disingenuous. elections scale linearly with precincts. In precinct count, every single vote is tallied in a verifiable way that night, before the pollworkers even finish cleaning up the poll stations. and all without computers.
@thegrimscalper@VladTheInflator 100 percent agree with this. the problem is when you actually trust the ai to be more than it is. but it can do way more, even today, than it does. proper process really is the key. but the same process you use to manage people fails with agentic.
@thegrimscalper@VladTheInflator yeah, it screws up ALOT! but so do people. the handle the people frailty with process, we are learning how to use process to handle ai fragility as well.
@thegrimscalper@VladTheInflator yeah, great for perl and stringy stuff. but for large codebases, agentic and then ai task delegation is the only way to go. the context management almost forces it. now, this is the thing that gates my eng velocity. as context increases, so does eng work chunk size and velocity.
@thegrimscalper@VladTheInflator i use it for architecture. example: give me 10 different methodologies for maintaining consistency in a distributed network. rank them by latency and fragility.
then that feeds other design prompts that generate design specs.
@thegrimscalper@VladTheInflator i just use different prompted ai and sometimes different models to do the planning stages. then the planners basically just commit the plans as md files in the repo. then code agents grab them and roll. i prompt mostly planners and testers.
@thegrimscalper@VladTheInflator i now have totally different metrics for measuring dev productivity. my expectations are now logarithmic accelerations of productivity and i really measure it.