# John H Boyer

- Researcher: John H Boyer
- Role: Visiting Assistant Professor of Philosophy, Loyola University New Orleans
- Field: philosophy
- Tools: Claude Code (Fable as orchestrator with Opus and Sonnet subagents); Codex (GPT-5.6); Grok 4.5; CLAUDE.md instruction files
- Published: 2026-11-20
- Source: https://johnhboyer.com/
- Written with the current flagship models (Claude Opus 5.5 and GPT-6.1 Sol) available.

## What they did

Boyer works on Aristotle. He writes on his site that for years he wanted what no one site offers: the TLG has the Greek texts, search and parsing but no translations, Perseus has translations but no good way to read them beside the Greek, and the Loeb has the facing page but no search or parsing. He wanted all of it in one place, so he built the Aristotle Reader, then Plato and Homer readers on the same model. The readers put Greek beside English with click-to-parse morphology, a Greek lexicon and search across the whole corpus. His site also lists a Greek Drama Reader covering all 44 plays and Grammata, a Greek and Latin grammar reference built on Smyth and on Allen and Greenough.

He writes: "I built the sites using AI but the translations are not AI: they are published, public-domain translations set beside the Greek." In the Homer Reader, he says the maps, scene summaries, timelines and genealogies "are drafted with AI assistance and then reviewed by hand, book by book". The site does not name the AI tools he used, and his public repositories do. At the conference, in reply to my question about hosting, Boyer told me that he hosts the sites on Cloudflare, having moved to it from GitHub; his files record the move from GitHub Pages to Cloudflare on 19 November 2026, the day before his talk.

His GitHub profile describes him as a "humble philosophy professor vibe coding projects for digital humanities". The repositories for the Aristotle, Plato and Homer readers are public, and each has a CLAUDE.md file of standing instructions for Claude Code. On 20 November 2026 the Aristotle repository had about 1,530 commits, Homer about 800 and Plato about 600. The Aristotle repository was created in June 2026 and the Homer repository in July.

The Homer instructions make the main Claude Code session, running the Fable model, an orchestrator that plans, writes briefs, integrates and commits. It hands the reading and writing to other agents: Sonnet by default, Opus for hard judgement including questions of Homeric philology, and Codex and Grok as implementers and reviewers, with at most five agents at once. Rules in the file are dated and attributed to him, for example "Vulgate lineation is sacred" (never renumber lines), translations on the site must be public domain under US rules, and "Never invent an identification". Certain steps are reserved for him: deploying, merging to the main branch, anything that costs money, copyright judgements, and changing a piece of apparatus from draft to reviewed.

One night in July 2026 he let the agents build unattended. The note for that night keeps a numbered list of decisions held for him and states that the gates (no deploy, no merge, nothing that costs money) were unchanged.

## What they learned

His site states no lessons as such, while his repository files record several. The site does show how he separates AI-drafted material from published sources. The translations are published ones, and the Homer Reader's About page tells readers to trust the underlying text and translations fully. Maps, scene summaries, timelines and genealogies are the parts drafted with AI assistance.

Until a book has been reviewed by hand, it carries a Draft badge wherever it appears. Readers are told to treat a badged summary, identification or family-tree link as provisional until the badge is gone. At the time of reading, the character apparatus had been reviewed. The instructions say the same: AI-drafted apparatus stays "draft" until he changes it.

Every place identification in the Homer Reader also carries a certainty tier: certain, traditional, speculative or mythical. An identification that comes from later tradition and not from the poems says so and names the tradition.

From 28 July 2026 the instructions require every Claude agent's work, including the orchestrator's own edits, to be reviewed by GPT or Grok, and every Codex implementation to get a checking pass from Claude before it is committed. The routing table records that Grok's content checks "twice found defect classes Sol/Opus/Claude all missed, with raw-line evidence"; Sol is a GPT model. An adversarial review by Codex dated 20 July 2026 listed, among other things, a capitalisation bug that corrupted the morphology of at least 711 word occurrences, and an "Epithets & formulas" feature that was not detecting formulas.

The Homer instructions include a failure-mode registry headed "append dated lessons — a lesson not written down will be repeated". One entry describes a check that compared label positions before and after a map change. It reported only one label moved, when four labels had been dropped altogether, and a Codex review caught it.

Hand-over notes are kept per track of work and rewritten at each hand-over. The Aristotle instructions forbid a single shared hand-over file because on 25 August 2026 one session overwrote another track's notes through that filename.

## Limits of this account

This account rests on Boyer's own project website and on parts of his public repositories. The website is undated and says nothing about how long the work took. The repository files are instructions and working notes written for and partly by the agents, so they show what was asked for and recorded. They do not show how consistently it was followed. I read parts of the instruction files and notes for the Homer, Aristotle and Plato readers, not the code or the commit history.

Boyer presented this work at the Ars Inquirendi conference in Oxford on 20 November 2026 under the title "Homer for Pleasure, Aristotle for Study: What Agentic Coding Makes Possible for Scholarly Digital Editions". This account draws on the talk only for that one answer: the public programme gives only its title and the recording is for registered participants. I have not checked the readers' texts, parsing or search for accuracy.
