AI case studies
Since these figures may be hard to believe at first, let us set out the premises up front. This is neither an estimate nor a theoretical value; it isa measured value tallied from the commit history of every Git repository and the actual development servers of a single development PC that our engineer really uses. The period is about 3.5 months from mid-April to the end of July 2026, and the work was carried out by one engineer and an AI coding agent (Claude Code).
| Metric | Measured value |
|---|---|
| Systems / projects developed or updated | 22 |
| Code created or updated (lines added) | About 246,000 lines |
| Number of commits | 446 |
| Period | About 3.5 months (most of it concentrated in the last 6 weeks) |
For the credibility of the figures, the tallyexcludeslibraries (node_modules, Python virtual environments, etc. — about 1,000,000 lines), lock files, binaries such as images, assets merely migrated from existing systems, and code duplicated across repositories. Conversely, since work built directly on servers is not included, the actual volume of work skews higher than this figure.
Done by hand, how many person-months?
Converted using the common productivity benchmark for business-system development, "1 person-month = 1,000–2,000 lines" (including design, implementation, testing, and debugging), 246,000 lines equates toabout 123–246 person-months — that is, 10–20 person-years. At a rate of 1,000,000 yen per person-month, that is a development volume on the order of 120–250 million yen.
One person handled that in 3.5 months while juggling other work. By simple arithmetic that is 35–70× productivity, and looking only at the last 6 weeks where the work was concentrated, it exceeds 100×. Of course, line count does not equal quality or value. But this scale of difference can no longer be captured by the word "efficiency."The agent of production has shifted from human hands to AI agents — the same structural change as when the industrial revolution replaced the craftsman's hands with machines is now happening in software development.
What did we build?
- 7 SaaS products for stores and businesses (social-post management, inventory management, sales analysis, a LINE membership platform, AI review responses, in-store background music, and a study site) — the largest was 176 commits and 41,000 lines. The inventory-management SaaS was 23,000 lines in 3 days.
- 4 corporate websites (including a multilingual site; all built as static sites without a CMS, with updates operated by AI)
- A full renewal of a 20-year-old Java system (Field Report ①）
- A phased-migration gateway for a core system running on an end-of-support OS (Field Report ②）
- A prototype for a new-platform architecture study (Field Report ③）
- plus internal tools, infrastructure configurations, and documentation
Let us be honest: AI is not magic.
It wouldn't be fair to tell only the success stories, so here is what actually happened. Data that had been corrupted 10 years ago could not be restored even by AI (you cannot invent lost information). We had an incident where parallel work rolled back code, and we recovered it from Git history. And more than once, AI's initial hypothesis was later overturned.
That is exactly why we run AI-driven development as a package with a methodology: validate every feature against real data, keep discipline in Git operations, and let humans make the strategic decisions.Not a tool, but a methodology.This is why we were able to keep delivering 246,000 lines over 3.5 months without incident.
Are you facing the same situation?
- You have a legacy system whose spec no one knows anymore
- The modernization estimate is tens of millions of yen and measured in years, leaving you stuck
- You are looking for an option other than "switch to SaaS and adapt your operations to it"
We offer legacy modernization as the AI Re: Platform (AIR Platform for short).
Analysis of your existing source code is free.The results are delivered as asimplified "AIR Report"summarizing an inventory of your technical debt and an initial "rebuild vs. wrap" strategy (conducted under an NDA).Get in touch today.
Contact form(please add a note saying "Requesting free analysis (AIR Report)") / info@flagship-ai.jp
Free online seminar"Legacy to AI — The industrial revolution of the IT industry, and a field report on reviving a 20-year-old Java system"planned (dates being arranged).Pre-registration opens soon — we will announce it on this blog.
Series: "Legacy to AI — Field Reports"
- Overview: 22 systems and 246,000 lines on a single PC — a measured record of 3.5 months of AI development (this article)
- Field Report ①: Fully modernizing 20-year-old Java in about a week
- Field Report ②: Adding a new entry point to a core system with zero lines of existing code changed
- Field Report ③: A month-scale architecture study, done with a 2-day prototype + real measurements
* The figures in this article are measured values as of the end of July 2026. Development outcomes vary depending on the state and requirements of the system.