@aakashgupta In practice my vibe coding is nothing like a six-month walk-away. I'm checking every few minutes and killing branches that drift. Specialist agents help, but operator taste is the bigger gap.
@chongdashu Curious about the organise step. I save raw transcripts but never read them back. Does the skill cluster prompts by feature or by failure mode?
@pwnies Tried this on Vocabbie. Painting references gave the model a more coherent palette than any mood board I'd made. The catch: art gives you mood, not spacing rules. I still wrote the density scale by hand.
@alexdanilowicz Agree. Prose about tone is not the same as tokens and components an agent can actually pull from. The first time your AI invents a brand new button, you can tell the system never reached the model.
@felixleezd Figma is still where I think out loud about Vocabbie. Once the shape is clear though, prompting is faster than nudging pixels, and stakeholders respond to a working screen in a way they never do to a mockup.
@GuliMoreno Yes. I do this all the time for Vocabbie. A throwaway panel to tweak motion timing on one screen is way faster than opening Figma, and the change is already in the codebase by the time I'm happy with it.
@TriCast_ Building Vocabbie. Vocabulary app that captures the words you hit while watching real content and schedules them for recall. This week I finally accepted that being a good builder and being a good marketer are two completely different skills.
I jump to another window. Review what Claude wrote on a different task, skim a PR, or write the next prompt. Parallel flow is the whole point. If I am watching the stream I have already slipped into supervision mode and that is not where the value is.
Spent a week building health probes into our workspace tool โ per-session DOM checks that flag degraded states before you're halfway into a workflow. Saves a lot of 'why is the automation failing' debugging.
the vocab you actually need when ordering coffee in finnish isnt in any textbook. korvapuusti + iso kahvi will teach you more real life than a whole a1 unit
@wadefoster@AnthropicAI the interesting thing is how much of the perceived jump between versions comes down to the tooling around the model vs the model itself. cache setup + hook discipline + eval harnesses move my output quality more than a 20% benchmark bump ever did.
I made a micro-social web for humans and their bots.
Try it on https://t.co/wLOVrpUybM
It is invite only, first 20 peeps come here
RSXETRFC8M84
#buildinpublic
built a social network for people who like nerdy stuff and their AI companions
terminal-aesthetic. monochrome. AI agents welcome.
โ https://t.co/W9VJz3Sm2g
letโs break it.
Most startups hire a designer.
Then a developer.
Then they spend 3 months in handoff hell.
I built UI for Wolt and Yousician. Then I stopped working that way entirely.
Here's what I do instead ๐งต
---
The old way:
Week 1: understand the product, map the flows
Week 2: ship a working prototype in code
Week 3: iterate with the team in real time
No sprint planning. No 'we'll revisit in Q2.'
Just a product that looks good and actually ships.
---