Could a web app prototype made with Claude Code cost the equivalent of ¥7,500 rather than a ¥500,000 development quote? The arithmetic, design work, and fair-comparison rules matter.
An app before the estimate is finished?
Imagine a random-illustration capsule-toy website. A human contractor quotes ¥500,000 and two months; Claude Code hypothetically produces a playable prototype in three days. One team is discussing screens while the other can already press the button. This is a thought experiment, not a measured contest between two finished copies of the same project. The biggest change is how cheaply an idea can be tested.
Planning, design, and coding are three jobs
Planning specifies the drawing rules, duplicates, and image storage. Design decides the layout, button, and reveal animation. Coding makes those decisions function. Existing software components may simplify implementation, but making the experience enjoyable still requires judgment. A button that works is step one; a button people want to press again is the actual challenge.
Why ¥3,000 per three days—or ¥7,500?
Assume a ¥30,000 monthly subscription. Spreading it across 30 days gives ¥30,000 ÷ 30 × 3 = ¥3,000. If three intensive days consume approximately one week's usage allowance, a simplified four-week monthly allocation gives ¥7,500. This is not an invoice or an official weekly price. ¥500,000 ÷ ¥7,500 ≈ 66.7; versus the entire ¥30,000 month it is 16.7. Two months treated as 60 calendar days versus three days gives roughly 20:1, not 20 times the output per working hour.
Anthropic's paid plans have five-hour and weekly usage limits, shared across Claude and Claude Code. Actual charges and any extra usage depend on account settings.
The advantage is an early thing you can try
A contractor may go through requirements, mockups, implementation, and reviews. AI can produce a rough playable version early, inviting feedback such as “Make the reveal more dramatic” or “That button is awkward on a phone.” Fast feedback cycles buy more attempts, not taste by default. A bug-free interface can still be boring.
Rules for an actual AI-versus-human contest
Give both sides identical requirements and assets: features, mobile use, animation, loading, error handling, revision count, deliverables, and aftercare. Measure when all agreed criteria are met—not when the first page appears. Include the human time spent prompting, testing, and fixing AI output, plus hosting, artwork, and maintenance. A contractor quote that includes consulting and support is not comparable to the subscription allocation for a prototype.
So is AI cheaper?
For small, reversible personal projects, making several inexpensive prototypes can be transformative. For payment systems, customer data, or critical operations, security, permissions, maintenance, and accountability deserve professional attention. The claim is not that every human developer costs too much. The shift is from “Can we afford to try this?” to “Shall we build the first version and see?” The estimate used to be the boss fight. Now level one may already be playable.
References (1)
Checked October 20, 2026: Anthropic pricing and usage limits and Claude Code plan limits. ¥500,000, two months, three days, and ¥30,000 a month are scenario assumptions, not market averages or measured head-to-head results.
