reality of a vibe coder:
it works perfectly on localhost.
then you get actual users.
and instantly discover why senior engineers exist.
the vibes do not scale.
harsh reality:
you iterate fast and criticize yourself daily.
but your boss has a massive ego.
they can't ingest info at your speed.
so they just block you.
build a startup.
#BuildInPublic#vibecoding
obsessing over the smartest llm is a trap.
you don't need expensive models.
downgrade your ai.
use strict linting and retry loops with explicit errors.
#BuildInPublic#vibecoding
reality of a vibe coder:
shipping fast is easy now.
but nobody can draw a basic funnel.
no mapped initiatives or feedback loops.
enjoy building for zero users.
#BuildInPublic
stop lying to yourself about deepseek/kimi/qwen.
you’re spending weeks on "optimization."
just pay the Anthropic tax and ship.
latency is the death of your startup.
#BuildInPublic
reality of a vibe coder:
you feed a random DESIGN.md to an agent.
you sit back and expect magic.
reality hits you with missing padding.
undefined components and broken animations everywhere.
#vibecoding#BuildInPublic
building for devs: "wow cool stack"
building for normies: "where do i click?"
coding is the easy part.
bridging the brain gap is the trap.
most of the builders aren't ready for that.
#BuildInPublic#Startup#Coding
you can ship the cleanest v0 in an hour.
but if you have zero visual assets?
the community literally won't see you.
lack of design is the ultimate distribution killer.
#Startup#BuildInPublic
the internet is a graveyard of perfect code.
genius builders making incredible products.
that literally nobody will ever use.
why do developers hate selling so much?
#BuildInPublic
the @GeminiApp default loop is exhausting:
ask a question.
get a weak answer from the fast model.
realize what happened, switch to pro, ask again.
does anyone actually prefer the fast default?
spent 4 hours architecting a video editing agent product
went through: trigger dev, fly machines, modal, e2b, cursor sdk, claude agent sdk
turns out the answer is just vercel supabase (1/6)
#BuildInPublic
the stack:
- vercel → next.js, workflows, ai sdk, chromium for export
- supabase → postgres (projects/compositions/messages), storage (assets/mp4s), auth, realtime
- plus any model provider (5/6)
takeaway:
- "AI agent infra" is a marketing category, not always an engineering need.
- for structured-content editing products, your stack is: web framework durable execution db storage.
- if one vendor covers half of those, your architecture just got simpler. (6/6)