Scholar KarlMuth Completes Japanese Securities Law Paper with Multi-Agent Pipeline
Scholar KarlMuth shared a detailed account of how he used a suite of custom AI agents to produce his paper on Japanese market regulation, "Fast and Furious Tokyo Grift: 30 Years Post-Milhaupt," set to appear in Volume 22 of UPenn's Asian Law Review. The pipeline spans five stages—source gathering, academic advising, language calibration, defense preparation, and citation formatting—offering a complete example of AI agents deeply embedded in legal research and writing.
Confirmed
- Source gathering: Two autonomous agents were built to explore Japanese newspaper archives and automatically compile a corpus of primary Japanese-language sources, freeing the work from dependence on specific search terms or strings; KarlMuth noted this kind of task would be extremely tedious even for bilingual research assistants or students
- Academic advisor: Pre-1996 market regulation materials, case law, and articles were isolated into a dedicated corpus, combined with key monographs, to create a 1996-era "clone" of Prof. Milhaupt, occasionally used as a sounding board for ideas
- Culture agent: Trained on Japanese laws and regulations plus local media coverage of securities enforcement cases, it calibrated the tone of descriptions and translations of Japanese phrases and terminology in the paper. KarlMuth explained the motivation: his Japanese tutor, though excellent, is not a securities lawyer; the agent generates tables listing every Japanese term in the manuscript, matching each term's specific usage in the author's own writing rather than its generic meaning
- Prep agent: To ready him for presenting and defending the near-final draft at a July 13 faculty workshop, a prep agent drew on the latest draft to quiz him every morning with three likely workshop questions, simulating a defense interrogation
- Citation formatting: A Bluebooking agent—used and iterated for two years—handled the final legal citation formatting
Why it matters
- This is an end-to-end multi-agent academic research case, demonstrating agents' potential to replace human expert assistants in tasks like processing non-English primary sources, "cloning" domain-expert advisors, and calibrating linguistic style
- The "culture agent" matches terminology to the author's own writing habits rather than generic definitions, exemplifying a practical approach to personalizing LLMs to an individual's style
2026-09-03 ~ 2026-09-03 · 12 related posts
Primary sources
- Researcher built AI agents to mine Japanese newspaper archives and clone a 1996 law professor — KarlMuth ·
- New paper surveys 30 years of Japanese market regulation since Milhaupt's classic article — KarlMuth ·
- How one writer used culture and Bluebooking agents to publish in UPenn's Asian Law Review — KarlMuth ·
- [source] New paper surveys 30 years of Japanese market regulation since Milhaupt's classic article — KarlMuth · 2026-09-03
- [source] Researcher built AI agents to mine Japanese newspaper archives and clone a 1996 law professor — KarlMuth · 2026-09-03
- Two agents mine Japanese newspaper archives to build a primary-source library — KarlMuth · 2026-09-03
- He quarantined pre-1996 sources to build a 'clone' of Prof. Milhaupt as a sounding board — KarlMuth · 2026-09-03
- A 'prep agent' quizzes him with three mock-defense questions every morning — KarlMuth · 2026-09-03
- Culture agent tables Japanese terms precisely; a two-year-old Bluebooking agent finishes the citations — KarlMuth · 2026-09-03
- A 'culture agent' fine-tunes Japanese legal phrasing in academic writing workflow — KarlMuth · 2026-09-03
- Scholar uses paired archive-exploring agents and a 'Prof. Milhaupt circa 1996' clone as research aides — KarlMuth · 2026-09-03
- Scholar builds a prep agent that quizzes him with three questions each morning before faculty defense — KarlMuth · 2026-09-03
- Scholar builds a culture agent trained on Japanese law and prosecution coverage to tune legal writing — KarlMuth · 2026-09-03