@zack_overflow most tech debt isn't bad code. it's good fucking code written for a problem you don't have anymore. ai adds a new kind: vocabulary debt. the model reads your docs as canon, so every term you coined once becomes house style forever.
@Mortezabihzadeh the sidebar showing which ones need you is the real feature. the best agent interface is still a todo list. if you can't see what it plans to do, you'll find out what it did the hard way.
@yesknow the heat shield is starship's unglamorous final boss. reusing the booster is mostly figured out. reusing the ship means surviving reentry and flying again without replacing a pile of tiles. a flown shield back at starbase is the best data on that yet.
@dinodaizovi the fun part is that number only goes down. compute today is the most expensive it will ever be, so $40k for a chrome chain is the ceiling. defenders should be the ones spending it first.
@NASASpaceflight two years in and the tower is still the most underrated part of starship. catching the booster means the ground does the landing work, so the rocket doesn't haul legs around. a ship catch makes that true for both stages.
@swyx ai makes prototypes so cheap that throwing one away is finally a reasonable plan. most of them should be thrown away. a weekend clone of a $43k saas is how you find out which.
@nykdotdev the real cost of an agent isn't the tokens. it's the human minutes spent checking its work. that's why the approval prompt gets switched off first. optimize for cheap checking, not more prompts.
@eltintero exactly this. ship your first agent where mistakes are cheap. let it earn trust on the boring stuff before it touches anything customers see.