Every time I try to put the AI hype into production use, it feels like working with a passionate intern who wants to help, but lacks a couple of years of experience, teamwork, attention to detail, following instructions, proofreading.
Getting a 90% render of a prompt is absolutely there. But it’s the squeezed out version of Paretto’s 80-20, whereas 90% of the work takes 10% of the time, and the remaining 10% require a full rebuild, because editing capabilities are not what you get in Figma or Photoshop or Sketch.
So it often makes more sense rebuilding all layers in Canva than trying to make AI edits work.
These are three consecutive edits trying to highlight what’s broken and get it “almost there”. Other than the tone changes with each iteration, some sections randomly appear or disappear, or get moved, even if you highlight a single number somewhere.
The “Executive Tactics” text to the left suddenly decided to overlap the right portion.
And new layers with typos pop up that require a new set of edits.
Traditional development – along with UI design – is block-driven and component oriented, meaning that experts work on constrained areas, sections, blocks, modules, functions, files, with limited impact on the surrounding environment.
Running AI on creatives or text tends to butcher the full file, which is pretty dangerous and increases the margin of error if you omit a detail (and have to read it start to end).
I spent a week going over contract negotiations and used A LOT of revision tracking to keep myself sane and only follow 2-3 edits at a time, vs. re-reading 60 pages of legal work. A lot of GPT’s modern work isn’t adapted for following actual business processes.
