An Anonymous Solo Developer's ERP Integration SaaS: ~¥3M MRR with Zero Employees, Zero Ad Spend — and No Hand-Written Code for Six Months

MRR of roughly ¥3,000,000. Zero employees, zero ad spend. And for the past six months, not a single line of code written by hand — a post published on Zenn in March 2026 by an individual developer was read for exactly this three-part hook. The subject is neither a flashy AI app nor a consumer service, but a proxy server that converts a legacy ERP’s “insane asynchronous SOAP interface in XML” into clean, stateless RESTful JSON, the plumbing of enterprise systems.

Let us state this upfront: this case is an anonymous self-report, neither the operator nor the service name is disclosed. The numbers come solely from the author’s own Zenn article, and there is no way to verify them externally. We cover it nonetheless because the article contributes something rare to the “AI writes the code” debate: real revenue figures attached to concrete operational detail. Readers should take every number below with that caveat.

The Published Numbers

The product’s origin goes back years, to a draining ERP integration project that led the author to carve out a conversion layer for developers with the same pain. Customers are developers worldwide, the article describes “a person on the U.S. West Coast” rushing in for an emergency fix in a legacy ERP environment. It is B2B, but there are no sales calls: it sells self-serve from $30 a month.

Acquisition: Staking Out Error Messages

The acquisition engine behind ¥3M MRR with zero ad spend is, at bottom, SEO, but the keywords are unusual. The author describes it as staking out “hopeless error stack traces.” Developers wrestling with legacy ERP do not search for product names. They search for the exact error message in front of them. So the author places technical posts on Dev.to and English-language blogs with error messages as their titles, funneling to the paid solution at the end.

What makes this design strong is the searcher’s extreme readiness to pay. For a company whose operations have stopped on an error, $30 a month is not even a decision. And the search surface of “specific legacy ERP errors” has almost no competing content, big, high-domain-authority media have no incentive to enter. Like the case that carved a $1M/year niche out of the Microsoft ecosystem, the structure is to pick up the “inconveniences nobody bothers with” around a giant platform.

Inside the “No Hand-Written Code” Operation

The biggest reason the article traveled is its detailed account of AI operations. In the author’s words, for the past six months he has “not spent a single second on traditional coding, opening an editor and typing by hand.” He invokes Claude Code from the terminal, hands over requirements and specs in prose, and lets it loop through test generation and automated fixes. Error monitoring is likewise handled by automated bots through analysis and repair.

Note what this is not: a story of AI building a business from zero. The product itself was built by the author years ago, and the market interface (which errors get searched, what customers pay for) was established through human trial and error. What AI replaced is maintenance and modification of a mature product with a settled spec: the work AI is best suited for. In contrast to Healthchecks, a one-man SaaS run by hand for a decade, which is an accumulation of its founder’s manual craft, this case should be read as a division of labor that hands over only post-maturity operations to AI.

The operating philosophy is consistent: the contact form is abolished, cancellation is one click. Human touchpoints are deliberately erased and the UI is optimized for user self-service. Zero-employee B2B works only on the premise that no human staffs support.

What to Discount

To repeat: this case’s greatest weakness is unverifiability. Anonymous and unnamed, there is no dashboard or third-party reporting to back the ¥3M MRR. That it was published in Zenn’s technical community, with an internally consistent stack and operational detail, is circumstantial support, not proof. Under this site’s editorial policy it counts as a primary (self-disclosed) source, but it warrants one notch less confidence than named cases with named services.

Structural risks are also legible in the author’s own account. On the market side, legacy ERP is slowly shrinking, not growing, and a conversion proxy loses its reason to exist the moment the vendor modernizes its API. On acquisition, single-channel SEO is fragile against algorithm shifts and the move to AI search. On tooling, an “AI writes all the code” operation imports a new dependency: the price and quality volatility of models and tools. The lightness of zero employees is inseparable from having no one to consult when these risks materialize.

What Generalizes, and What Doesn’t

Strip out the anonymity and the blueprint still reads. A B2B pain-driven niche, specific enough that the buyer searches by error message, pairs best with zero-ads, zero-sales self-serve. AI code generation works best not on greenfield development but on maintaining a mature product with a settled spec. Support-free design (self-contained UI, one-click cancellation) is a precondition of solo B2B. It is a textbook instance of the “developers selling to developers” pattern, kin to the case of growing a developer tool into a $50K MRR lifestyle business.

The limits are equally clear: this pattern is open only to those holding the entry asset of “an enterprise-system pain I suffered myself years ago,” because the resolution of that pain determines both niche selection and the persuasiveness of the SEO posts. And since the case is anonymous, the figure itself should not be a reproduction target. What to take away is not the ¥3M, but the blueprint: pain specificity, self-serve, AI-run maintenance.