{"slug": "show-hn-adrs-for-ai-deliberation-records-for-human-ai-work-versioned-in-git", "title": "Show HN: ADRs for AI – Deliberation records for human–AI work, versioned in Git", "summary": "A Show HN project called ADRs4AI proposes recording human–AI collaboration decisions as Architecture Decision Records versioned in Git, so the reasoning behind each choice persists rather than disappearing after a session. The method, demonstrated on the public frontier-model-observatory repository, has an AI stake open questions with options, rationale, confidence and a falsifier, while the human supplies a single ruling — in the cited example, GPT 5.6 Sol recommended a \"C — Bitemporal hybrid\" temporal policy and Jérémie Lumbroso answered \"C — Bitemporal hybrid\" from his phone. The workflow starts from a GitHub template repository, commits a record before work begins, and keeps open questions findable with the command grep -rn '^### QST' docs/adr/.", "body_md": "Your thinking stops disappearing. Your decisions start compounding.\n\nThe most valuable byproduct of the collaboration between a human and an AI\nis the thinking — and in ordinary practice, all of it evaporates. You can already\nmake good decisions with your AI. Every session produces reasoning you will\nneed again, and lose by default. **This keeps the why, so it compounds.**\n\nThis is the method’s product — one question from a live, public project’s own record, exactly as it sits in the repository:\n\n```\n### QST-HISTORICAL-TRUTH: How should corrections preserve what was previously believed?\n- Status: answered\n- Why asking: Conversation attribution needs “what was true or documented as\n  of date X,” while later research may correct earlier records.\n- Need: Approve or revise the proposed temporal policy.\n\n**Options**:\n\n- **A — Mutable present truth**: Correct records in place; rely on Git\n  history for prior states.\n- **B — Append-only assertions**: Never change claims; append corrections\n  and supersession edges.\n- **C — Bitemporal hybrid**: Correct descriptive records when needed, while\n  claims/events carry valid-time, observed-time, and explicit supersession;\n  Git remains a third audit layer.\n\n**Recommendation**: (by GPT 5.6 Sol at Perplexity Computer)\n\n**C — Bitemporal hybrid.**\n\n**Rationale**: Git history records when repository bytes changed, not when a\nrelease became valid or when a provider made a statement. Explicit\nvalid-time and observed-time support historical queries, while supersession\nedges preserve corrections without forcing every typo into an eternal\ndomain event.\n\n**Confidence**: High that Git history alone is insufficient; medium-high\nthat two temporal axes cover the first release.\n\n**Falsifier**: If real attribution cases require additional independent\ntimes, such as product-availability time distinct from announced and\neffective time, the temporal model must expand rather than overload either\naxis.\n\n**ANS:** (by Jérémie \"Sonnet 4.5\" Lumbroso (via ADRs4AI mobile))\nC — Bitemporal hybrid.\n```\n\nFrom [`frontier-model-observatory`](https://github.com/jlumbroso/frontier-model-observatory/blob/main/docs/adr/0003-epistemic-records-and-generated-views.md),\nverbatim. The AI staked the question; the recommendation — options,\nrationale, confidence, and what would prove it wrong — is GPT 5.6 Sol’s;\nthe one-line ruling came from the founder’s phone. **That’s the whole\nmechanic**: The AI carries the reasoning, the human spends one judgment —\naccept, override, or [any answer in any shape](https://github.com/jlumbroso/frontier-model-observatory/blob/main/docs/adr/0007-legible-continuous-integration-and-release-artifacts.md) —\nand the record keeps the why, for every mind that comes later.\n\nReal commands, the flow the founder actually runs — one seed, one question, one ruling:\n\n```\n# 1 · Your copy of the template\ngh repo create my-project --private --clone \\\n  --template ADRs4AI/human-ai-collaboration-template-A\ncd my-project\n\n# 2 · Open CLAUDE.md, fill in the top block (messy is fine),\n#     then stake the record before any work starts:\ngit add -A && git commit -m \"chore: the record begins before the building does\"\n\n# 3 · Recruit — open your agent and offer the mission:\nclaude    # or your agent of choice\n#   \"Hello — I have a mission for you. Read about it and tell me\n#    if you accept, or any pushback.\"\n```\n\nThen dump your thinking into a first **seed** — dictation is fine; your AI\nchunks it into ADRs that keep your actual words, the options, the reasoning,\nthe open questions. Answer the first question it stakes. Every open question\nis one grep away, always:\n\n```\ngrep -rn '^### QST' docs/adr/\n```\n\nOne seed, one question, one answered ruling — that’s the loop, and it already compounds.\n\nRunning a crew — several seats, wake-on-brief, you stop being the message\nbus — is the *optional* next chapter: [steps 7–9 of Get started](https://adrs.systems/get-started/).\n\nDump your thinking into a **seed** — messy is fine, dictation is fine; capture\nfirst. Your AI chunks it into **ADRs** that keep what usually evaporates: your\nactual words, the options considered, the reasoning, the open questions. Every\nquestion is findable by one grep — `### QST:` — or surfaced by the editor\nextension, the MCP server, the phone in your pocket; and each arrives with the\nAI’s options, its recommendation, its rationale, its stated confidence,\nand what would prove it wrong — so answering streamlines to *accept or\noverride*: Cognitive load down, bandwidth freed for the deliberations that\nmatter. It’s where conversations become decisions that keep their reasons —\n**deliberative programming**, in one habit.\n\nYou don’t need to be an experienced engineer to start. The method was built\nthrough daily practice by one human and a growing crew of AI collaborators,\nand its whole point is that **the record does the remembering** — as much\nfor the veteran with too many projects, as for the beginner building their\nvery first one.\n\n*The extension’s Open Questions view, on this project’s own record: A question, its options, an attributed recommendation — and the answer box. Question 16 of 75, that day.*\n\nNot transcript archival. Not vector-database “memory.” Not an orchestration platform. Not a claim that an AI possesses persistent identity — the record holds, minds change, and the work resumes either way.\n\nThe most valuable thing that happens between a human and an AI is not the\ncode, the draft, or the answer. It’s the *thinking* — and in ordinary\npractice, all of it evaporates. […] It does not make an AI persistent, and\nit does not make anyone remember. **It makes the work resumable — and the\nthinking compound.**\n\n— from [*What is ADRs4AI?*](https://adrs.systems/what-is/), the founding essay. A stranger can\nread it cold; its authorship and review record are public, because that is\nthe method.", "url": "https://wpnews.pro/news/show-hn-adrs-for-ai-deliberation-records-for-human-ai-work-versioned-in-git", "canonical_source": "https://adrs.systems", "published_at": "2026-10-25 00:43:47+00:00", "updated_at": "2026-10-25 01:12:35.212552+00:00", "lang": "en", "topics": ["ai-tools", "developer-tools", "ai-agents", "artificial-intelligence"], "entities": ["ADRs4AI", "Git", "GitHub", "GPT 5.6 Sol", "Perplexity Computer", "Jérémie Lumbroso", "frontier-model-observatory", "Claude"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/show-hn-adrs-for-ai-deliberation-records-for-human-ai-work-versioned-in-git", "markdown": "https://wpnews.pro/news/show-hn-adrs-for-ai-deliberation-records-for-human-ai-work-versioned-in-git.md", "text": "https://wpnews.pro/news/show-hn-adrs-for-ai-deliberation-records-for-human-ai-work-versioned-in-git.txt", "jsonld": "https://wpnews.pro/news/show-hn-adrs-for-ai-deliberation-records-for-human-ai-work-versioned-in-git.jsonld"}}