Alexander Marquardt runs a coding-agent fleet that merged 1,652 pull requests in a month on an e-commerce search solution, with 0.2 to 0.7% later reverted
Alexander Marquardt describes an e-commerce query interpretation layer built on Elasticsearch that grew from a 4,000-line query rewriter into a full search solution of about 620,000 lines, built with extensive use of coding agents. In October 2026 its main repository merged 1,652 pull requests with a median size of 492 changed lines; across four repositories the project merges about 690 a week. Agents work in their own clones under a control plane that allocates a fixed number of slots by weighted lanes, with a 15-minute watchdog. Every PR is reviewed by a separate adversarial review agent and must pass 136 CI jobs. Tests are 70% of the code and run on recorded data, and every written rule cites the incident that produced it.
The numbers (as reported)
| Merged PRs, main repo, October 2026 | 1,652 (about 373 a week), median 492 changed lines |
|---|---|
| Merged PRs, all four repos | 2,760 in ISO weeks 41 to 44 (about 690 a week) |
| Merged PRs, main repo, April to 8 November 2026 | 6,234 |
| Solution code | About 620,000 lines (backend and frontend) |
| Tests | 2.1 million lines; 27,635 backend test functions and 12,467 frontend test cases |
| CI | 58 workflows, 136 jobs |
| Merges later reverted | 0.2 to 0.7% |
| Token efficiency | 3.3x fewer tokens per changed line; 4.4x more PRs merged a week on about the same spend |
Facts
- Company / subject
- Alexander Marquardt (e-commerce search solution)
- Tools
- Claude Code
- Codebase
- mixed
- Work type
- New product
- Stack
- Python, TypeScript, Elasticsearch
- Published
- 2026-11-08
- Last checked
- 2026-11-13
What the source claims Practitioner (own blog/paper)
Claimed by the builder, writing about their own work. Not checked independently.
- 1,652 pull requests merged in the main repository in October 2026 1,652 merged PRs
- Only 0.2 to 0.7% of merges later reverted 0.2 to 0.7%
- Each changed line takes less than a third of the tokens it did, with 4.4 times as many PRs merged a week on about the same spend 3.3x drop in tokens per changed line; 4.4x PRs a week
Caveats
Self-reported by the engineer who runs the project, on his personal site; the employer and team size are not stated. Merged PR counts and line counts are throughput measures the author himself says are imperfect. The author says rework is about one fix round per merged PR, and he does not yet measure how many real shopper queries are interpreted correctly. Claude Code is named only in connection with hooks; the full set of agents and models is not stated.
More stories by tool: Claude Code
Related topics: New products built with agents