1. Designing a universe is much faster than shipping one
The idea was wonderfully silly: let people draw directly on a website and drop their pictures into a giant shared capsule-toy machine. Each picture can be won once, then disappears from the prize pool. The winner keeps it in a collection and can share it online. The artist might suddenly spot their five-second doodle and shout, “Someone actually got my drawing!”
As drawings accumulate, the machine expands from room-sized to building-sized, then Earth, solar system, galaxy, and universe. A cosmic machine spitting out a badly drawn circle is the entire punchline.
The brief also gained ads, protection against spam, storage limits, deletion priorities, psychological and marketing research, design mockups, video, implementation, tests, and deployment. That instruction set took roughly 22 minutes. What was finished was the brief, not the product. Still, getting from a weird thought to an actionable assignment that fast is a big change.
2. The cute drawing game hides a serious backend
A drawing canvas is relatively straightforward. “One picture, one winner worldwide” is where engineering becomes tricky.
What if two people win the same picture at once? What if the global gacha and an artist's machine both claim the same item? What if the connection drops mid-draw? What if a bot uploads 100,000 identical doodles? What should be deleted when storage fills up?
A public service needs consistent shared inventory, duplicate-draw prevention, reports, cleanup, collections, sharing, ads, and monitoring.
The free-only infrastructure requirement makes matters more interesting. Cloudflare Workers Free allows 100,000 requests per day; its free D1 database enforces daily read and write limits.[3][4] Keeping costs at zero means limiting usage before those ceilings are reached.
Let the gacha grow to the universe. Do not let the invoice do the same.
3. How many days and how many yen?
A clickable demo and a maintained public web service have very different costs. These are illustrative project estimates, not actual contractor quotes.
| Scope | Rough calendar time | Hypothetical budget |
|---|---|---|
| Simple canvas and draw demo | Days to weeks | Tens to hundreds of thousands of yen |
| Small public version with posting, drawing and storage | 1–3 months | ¥1–5 million |
| Cosmic animations, social sharing, ads, reporting and operations | About 3–6 months | ¥5–15 million |
A larger build may involve four to seven people across planning, interface design, frontend, backend, testing, and release—not necessarily full-time for the whole period.
For context, Findy's 2026 survey put the average monthly freelance engineer contract rate around ¥808,000.[1] An app-industry guide similarly shows a broad span from prototypes to mid-sized products.[2] It's tempting to say “That's outrageous!” until you notice that the supposedly tiny gacha has quietly hired an entire miniature software company.
4. What if someone says, “I'll build it for ¥1,000”?
There was a funny idea: ask someone interested in becoming a web designer, purely out of curiosity, “How many days would this take, and how much would you charge?”
One question. Not a procurement questionnaire. Nobody agreed to bid on the entire universe.
But imagine the reply: “I'll do it for ¥1,000!”
A galactic space program for a thousand yen. Suddenly aerospace development seems affordable.
At that price a practice screen or tiny prototype is more realistic than a reliable, maintained platform. If both sides knowingly agree, it could be a fun project. Just settle scope, acceptance criteria, revisions, and responsibility before turning a joke into unlimited free support.
5. Claude Code versus a human would be fascinating
Give the same specification to an AI-led builder and a human-led builder, then see what happens.
The AI might generate the inventory logic quickly. A human designer might realize the gacha reveal is boring and invent a delightful animation. Either might surprise you. Visual quality and never awarding a picture twice are different skills.
A fair comparison aligns scope, deadline, and launch conditions, then evaluates working products: build time, money, fun, mobile usability, errors, double-draw prevention, and maintenance.
Research doesn't guarantee an AI win. In a 2025 METR experiment, experienced developers took an average 19% longer on certain tasks with the AI tools tested at that time. METR's 2026 update described participant-selection problems that made new speedup estimates difficult to interpret.[5] Neither result predicts this particular game's schedule.
“An AI could finish in half a day to two days” is a hypothesis until a real build passes real tests.
6. Not writing code doesn't mean you can't direct a sophisticated site
Even a personal article site can have search, multiple languages, interactive tools, publishing systems, and outage checks. The visible site looks like a blog; its operations look like software infrastructure.
Someone who orchestrates all that through AI instructions uses different abilities from someone who personally designs and codes the pages. Neither can be judged solely by how many lines they typed.
Product judgment, delightful interaction, testing, fixing failures, and keeping a service alive are all part of web development. A giant repository doesn't automatically mean a great product. A beautiful screen doesn't automatically mean a safe one.
7. The best outcome isn't a winner
A 22-minute brainstorm about a cosmic gacha somehow turned into a discussion of freelance pricing, AI versus human skill, and what “being able to build a website” even means.
The interesting question is what each builder protects, cuts, or invents when handed the same ridiculous assignment.
Maybe the human makes a surprisingly good design. Maybe the AI ships a working universe overnight. Both would be worth seeing.
And if someone really quotes ¥1,000 for the entire galaxy, check the scope first. The most science-fictional artifact may be the invoice.
Sources
[1] Findy 2026 freelance survey: https://findy.co.jp/3922/ [2] DearOne app development cost guide: https://moduleapps.com/mobile-marketing/app-dev-cost/ [3] Cloudflare Workers official limits: https://developers.cloudflare.com/workers/platform/limits/ [4] Cloudflare D1 free tier enforcement: https://developers.cloudflare.com/changelog/post/2026-09-01-d1-free-tier-limit-enforcement/ [5] METR 2026 productivity update: https://metr.org/blog/2026-02-24-uplift-update/
Locale: en
