completely unprompted wow moment from today - asked @DevinAI to make us a @tbpn style breaking news style announcement card for our AIEWF speakers drop tmr, FULLY expecting it to fail at a heavily visual task
and it oneshotted the WHOLE DAMN THING
@theo Honestly just use Devin. It's really really good now
Positive users praise Devin AI's rapid progress on one-shot visual design tasks as surpassing old limitations, while negative users report repeated crashes on simple tasks.
No Digg Deeper questions have been answered for this story yet.
play with it guys
https://www.ai.engineer/worldsfair/2026/speaker-announcement-trading-card
esp if youre an @aiDotEngineer speaker. looking forward to host you soon
completely unprompted wow moment from today - asked @DevinAI to make us a @tbpn style breaking news style announcement card for our AIEWF speakers drop tmr, FULLY expecting it to fail at a heavily visual task
and it oneshotted the WHOLE DAMN THING
@swyx @DevinAI @tbpn Devin Desktop has been crashing on my simple task for like 3 times in a row... how can I report this to someone?
@DevinAI ?
@swyx @DevinAI @tbpn which model did you use? I have Devin as a former Windsurf customer, but haven't found use of it still. Using Claude Code mostly
@altryne @swyx @DevinAI @tbpn Oh no! Can you share diagnostics?
@DevinAI @tbpn
🚨 BREAKING
(are we doing X right?)
@altryne @DevinAI @tbpn bug @morgantepell about it
@AlxTurovski @DevinAI @tbpn devin base model
@swyx @DevinAI @tbpn I think it was a piece of cake for her.
@swyx @DevinAI @tbpn Devin AI something new for myself ngl
@swyx @DevinAI @tbpn devin doing visual work better than humans is the part everyone's gonna be weird about quietly for the next few months but nobody wants to say it yet
@swyx @DevinAI @tbpn everyone figured design was the failure point.
nope. that gap between expectation and reality is the story. what else are we wrong about?
@swyx @DevinAI @tbpn It works because it has priors and the tbpn is slop-esque. I'm sure with a little work you can elevate it.
@morgantepell @swyx @DevinAI @tbpn Hey will DM
@swyx @DevinAI @tbpn Those moments are wild, though the real shift is realizing your old mental model of what it can’t do is already outdated.
@swyx @DevinAI @tbpn how much of debugging is just the gap between 'worked in the lab' and 'broke in the real world'?
@swyx @DevinAI @tbpn the visual oneshotting is the real signal. six months ago every coding agent choked on design tasks. that gap closed quietly
@swyx @DevinAI @tbpn now everyones gonna assume every visual task is trivial until it isnt
@swyx @DevinAI @tbpn The "fully expecting it to fail" part is the real tell. I run a few agents every day and visual work was the last thing I trusted them with. That wall came down so quietly most people were still arguing about benchmarks when it happened.
@swyx @DevinAI @tbpn Devin’s visual-gen ability is scaling faster than people realize — this isn't just a "one-shot flex", it's a leading indicator for agentic multimodal tasks that used to require custom fine-tuning.
@swyx @DevinAI @tbpn the pattern that gets me: agents nail the task i was sure they'd flub, then trip on the boring structured one i assumed was safe. my prior for what's hard for them is useless now. wild it oneshotted a visual layout though.