I am going through the process of redesigning the SlideMagic blog and have AI agents ‘working for me’ to do the heavy lifting on the coding side. The process showed me something interesting.
I can look at a web page and instantly come up with dozens, dozens of bigger and smaller tweaks that are totally obvious to me but not to my agents. On top of it, each takes time to implement, and might break something else in the design.
We have cracked AI for text, not yet for visuals. In 2026, AI models are still processing images via text as a shortcut. This is why AI stumbles. A first shot might look OK, but I have not seen AI design solutions that are good iterating and making tweaks.
Image By Rama - Own work, CC BY-SA 2.0 fr, https://commons.wikimedia.org/w/index.php?curid=906988