March 15, 2026
Vibes vs deadlines
100 hour gap between a vibecoded prototype and a working product
Internet dunks on “30‑minute apps” as devs say the last 20% is a nightmare
TLDR: A builder vibecoded a dino photo app: prototype in an hour, nearly 100 more hours to make it real. Commenters roasted “30‑minute apps,” rallied around the 80/20 rule, and split between “polish is hard” and “just build for yourself,” proving AI speeds starts—but not finishes.
A founder tried to “vibe code” a cute dinosaur profile‑pic app and learned the hard truth: you can get a demo in an hour, but shipping a real product eats your weekend. Vibecoding—letting AI write most of the code while you wing it—worked fast for a prototype, then slammed into design tweaks, mobile bugs, and weird image results. Cue the comment section chaos.
The spiciest crowd says the hustle-bro myth is over. One skeptic snarked that these “30‑minute apps” aren’t apps at all—just demo bait—and joked it still takes months to make a decent note app. The faithful showed up too, waving the classic 80/20 rule: AI rockets you to 80%, then the last 20% devours 80% of the time. Others argued the pain was self‑inflicted: if you’ve got fussy taste and can’t explain it clearly, AI will churn out Franken‑UI forever.
Then came the curveball: some builders bragged that vibecoding makes entire products unnecessary—why buy when you can whip up your own for yourself? A wholesome take compared AI agents to woodworking jigs: you still build the furniture, you just make better jigs. Meme check: “dinos went extinct, so did the vibe,” “Figma saves marriages,” and “ship > vibes.” The verdict? Vibes are fun; polish is pain.
Key Points
- •The author spent about 100 hours building a small app with AI-assisted “vibecoding,” finding a large gap between quick prototyping and a polished product.
- •They previously adopted LLM coding in 2023 at a startup, progressing from simple to complex features using tools like GPT-4 preview, ChatGPT, Cursor, and Claude Code.
- •For the new app, “Cryptosaurus,” they chose to manage infrastructure directly, avoiding platforms like Replit or Lovable to learn the full process.
- •Using ChatGPT to refine scope and Opus 4.5 (Plan Mode) for planning, and integrating Gemini API keys, they produced a working prototype in the first hour.
- •Most time was spent on design and robustness: iterating UI via LLMs, fixing front-end/mobile issues, selecting colors with Coolors, and addressing inconsistent outputs on edge-case profile pictures.