
making models design better @kritapp • I’m gonna be king of the Pirates.
Joined June 2014
- Tweets1.4K
- Following969
- Followers1.1K
- Likes5.6K
Replying to @lobanovskiy
100%. you can ask your agent to do a review of how your workflow and tool usage changed over time and it will make it obvious to stay flexible
teach your ai rather than handing them books and hoping
Replying to @designertom @elayadesigns
wont work in the long run, skills are suggestions, they bloat up context and mess up decisions when you have multiple of these.
make your ai learn skills, like how we humans do. give it tasks and once you’re happy walk back and turn that into skills. this is different now cause its also memory
Replying to @designertom @elayadesigns
wont work in the long run, skills are suggestions, they bloat up context and mess up decisions when you have multiple of these.
make your ai learn skills, like how we humans do. give it tasks and once you’re happy walk back and turn that into skills. this is different now cause its also memory
Replying to @jakubkrehel
we need something in the middle cause those flashbangs be really bad
Replying to @desdivj
how'd you build this? looks dope
video gen + code?
Replying to @edral_studio
are you pointing at self updating site?
i'm experimenting something like that for hydradb.com (not saas)
our engg keep hitting new benchmarks, we ran so many pricing experiments so our docs/blogs were always behind
fix is simple, components
Replying to @shrirajpawar04
buddy, server side was fucked
if 3 agent worked in same file it would crash, my droplet was not able to handle any more than 300mb
did simple stuff but major gains for sure.
vibecoding skill up 📈
spent 2 days making krit fast. no new features
undo: 9.3s → 5ms
style edit shows up: 8.9s → 4ms
moving a board saves: 1.9s → 71ms
projects and pages are loaded properly
fixed memory leakages
server memory: cut in half
learned a bunch of server related stuff doing this
Replying to @HarshaHegde0 @sankalpdomore
fastest way would be deploy on vercel and use their comments feature for preview deployments or claude artifacts
Replying to @MichaelFilipiuk
same thing works one level up with agents. i keep a list of every correction i make to claude output and it reads that before building. the list ended up carrying more taste than the components
but remember, agent reads these as suggestions so there still will be misses
Replying to @jamesm
this is cool, you can also ask it to create a canvas for your project. makes life easier to explore variations
another trick you can use is have stage of individual elements so focus on one part of the UI, that way you can iterate, refine each piece and reach "craft"
finding reference 101
searching "fintech dashboard" is how you end up with the same look as every tool
reframe the job before searching
"Exceptions screen" is a bad query. The job is "triage a queue every morning and clear it fast", so the best references might come from other domains
1. email triage (Superhuman)
2. code review diffs (GitHub showing what differs)
3. airline ops or support queues
and remember, pick for the reason, not the surface



