{"slug": "reviewing-patrick-collison-s-ask-for-an-llm-workflow-tool", "title": "Reviewing Patrick Collison's Ask for an LLM Workflow Tool", "summary": "Patrick Collison's description of an ideal LLM workflow tool, characterized as \"GNU Autotools × Notion,\" directly aligns with the Verivus Assurance Platform (vap) and its open-source verivus-oss component, according to a developer working on the project. The agentassurance component implements a \"living theatrical production studio\" that addresses Collison's identified friction points—semantic state persistence, mixed human-agent coordination, evidence traceability, and explicit governance—through a zoomable Production hierarchy, DAG-TOML-based intent declarations, and provenance-backed ijbCRUD operations. Real-world runs, including public X threads and a ledger distribution review governed by a 22-unit DAG-TOML plan, demonstrate the system's ability to produce shareable, versioned artifacts with cryptographic evidence trails.", "body_md": "Patrick Collison ([https://x.com/patrickc](https://x.com/patrickc)) recently outlined the [LLM workflow tool](https://x.com/patrickc/status/2063337800209179029?s=20) he actually wants.\n\nI know pointing at my own work can read as self-promotion. I'm actually trying to stress test the production model I've been running under the vap umbrella in verivus-oss.\n\nIt lands right in that gap (and the evidence from real runs, including public X threads and the recent ledger distribution review, is there).\n\nPatrick wants:\n\nHe summarised the desired feeling as \"GNU Autotools × Notion\", a system for a body of material that you want to process iteratively, where certain artifacts are important enough to preserve, version, govern, and reason about across time.\n\nThe diagnosis is accurate. For many of us the generation bottleneck has moved. The dominant remaining problems are semantic state that survives many iterations and participants, coordination that doesn't collapse under mixed human and agent work, evidence that actually travels with the work, and governance that keeps intent explicit rather than dissolving into chat history or ad-hoc folders.\n\nvap is the Verivus Assurance Platform, the umbrella under which the open verivus-oss work sits (and under which the deeper substrate in verivusai-labs is being built). The part that directly answers Patrick's friction is the living theatrical production studio, implemented as the agentassurance component.\n\nEvery body of work (a product, an initiative, even a single X reply series) becomes a zoomable Production inside the studio. The layout is the interface:\n\nProductions live in the left sidebar as the hierarchy. I can sit at the full Verivus portfolio level or zoom down to a 22-unit DAG-TOML remediation plan. The same rules and ijbCRUD pane apply at every zoom.\n\nWorkspaces fill the centre: Storyboard for the typed DAGs that hold intent declarations, depends_on and blocks relations, acceptance criteria, and evidence requirements as first-class versionable artifacts; Scene for the current focused rehearsal; Explore for semantic cartography; Working On for the live messy iteration surface.\n\nExhibition sits in the right sidebar: the compiled outputs worth preserving and sharing, carrying full chain of custody.\n\nShared Resources run along the bottom (Props, Cast, Timeline), with Next in Line holding the queued pipeline.\n\nThe central operating verb across every layer is ijbCRUD, provenance-aware and evidence-backed by construction. Closure roots travel with the artifacts. Assertions live in the canon. This is what makes the state survive iterations and participants instead of collapsing back into chat or untrusted folders.\n\nThis is Autotools × Notion lifted into a full production process, grounded in DAG-TOML plus the Agent Assurance specification. Explicit intent, evidence via closure roots, cryptographic provenance, IJB assertions as substrate, runtime-neutral by design.\n\nI've already run this evolving model in public: X replies and crossposts with multi-LLM consensus and evidence traces, the ledger distribution review governed by a living 22-unit DAG-TOML plan.\n\nThe plan and its evidence became the shareable Exhibition record. sqry itself has been used in real audits. Earlier articles and repo briefs on dag-toml and the production model are out there too.\n\nThese are real, usable artifacts.\n\nThe fuller vision lives in the internal verivusai-labs work under vap: the complete substrate (ijb, vault, ledger, integrity, meter and related crates), deeper studio refinements, and day-to-day use on larger efforts. I'm surfacing pieces as they stabilise. The published verivus-oss artifacts are the current on-ramp. This is nights and weekends alongside the day job, completely disconnected, with learnings feeding one way only (#ihaveadayjob).\n\nIf Patrick's description matches the friction you feel doing serious long-running agentic work, context that survives iteration, workflows that are versioned and governed, agents deliberately cast, outputs that can be exhibited with real provenance, this is the direction under vap in verivus-oss.\n\nThe published artifacts are the on-ramp.\n\nConcrete experiments (running sqry on a real stack, authoring a typed DAG, using the pipeline for output) and precise evidence-based feedback on what would make you want to direct or act in a real Production are especially welcome.\n\nRepo links and contact in profile. Early collaborators willing to engage the ontology and run real Productions are welcome.\n\nThe underlying conviction is that tools of this kind function as cognitive co-processors, common grace that removes a significant portion of the grinding burden of semantic entropy and coordination so the remaining human work can be higher-order direction and faithful stewardship of Productions.", "url": "https://wpnews.pro/news/reviewing-patrick-collison-s-ask-for-an-llm-workflow-tool", "canonical_source": "https://dev.to/wernerk_au/reviewing-patrick-collisons-ask-for-an-llm-workflow-tool-1odk", "published_at": "2026-06-07 01:21:41+00:00", "updated_at": "2026-06-07 01:41:39.884481+00:00", "lang": "en", "topics": ["ai-tools", "ai-agents", "ai-products", "ai-infrastructure", "large-language-models"], "entities": ["Patrick Collison", "Verivus", "vap", "verivus-oss", "verivusai-labs", "agentassurance", "GNU Autotools", "Notion"], "alternates": {"html": "https://wpnews.pro/news/reviewing-patrick-collison-s-ask-for-an-llm-workflow-tool", "markdown": "https://wpnews.pro/news/reviewing-patrick-collison-s-ask-for-an-llm-workflow-tool.md", "text": "https://wpnews.pro/news/reviewing-patrick-collison-s-ask-for-an-llm-workflow-tool.txt", "jsonld": "https://wpnews.pro/news/reviewing-patrick-collison-s-ask-for-an-llm-workflow-tool.jsonld"}}