Patrick Collison wants an LLM workflow tool, and he just described exactly what's broken in how most of us are currently shipping AI-assisted work. His recent outline on X hit the diagnosis right on the head: the generation bottleneck has moved. The real problems now are semantic state that survives multiple 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.

The GNU Autotools × Notion Vision

Collison summarized his desired feeling as "GNU Autotools × Notion"—a system for a body of material you want to process iteratively, where certain artifacts are important enough to preserve, version, govern, and reason about across time. This isn't just another chat wrapper or prompt manager. He's describing something that functions as a cognitive co-processor: semantic memory that survives iterations, workflows that are versioned and governed, agents deliberately cast for specific roles, outputs exhibit-able with real provenance.

What Collison Actually Wants

The requirements break down into five key capabilities: first, ability to manage input files (Markdown or similar) plus general-purpose context; second, real-time collaboration with snapshots or VCS integration; third, the ability to create and manage inference workflows and a stored set of prompts; fourth, access to general-purpose coding agents—not just chat models; fifth, some concept of compiled outputs or inference results that can be shared externally. These aren't revolutionary asks individually, but treating them as an integrated production system rather than disconnected tools is where most solutions fall apart.

Enter verivus-oss and the VAP Stack

Developer wernerk_au has been building exactly this under the Verivus Assurance Platform (VAP) umbrella in the open-source verivus-oss repository. The living production studio, implemented as the agentassurance component, treats every body of work—a product, an initiative, even a single X reply series—as a zoomable Production inside the studio. The layout is the interface: Productions live in the left sidebar hierarchy, Workspaces fill the center (Storyboard for typed DAGs holding intent declarations, depends_on relations, acceptance criteria; Scene for focused rehearsal; Explore for semantic cartography), Exhibition sits on the right for compiled outputs worth preserving with full chain of custody.

Breaking Down the Component Architecture

The stack maps directly to Collison's requirements: sqry handles the semantic/living graph and memory layer (input files plus context); weave provides CRDT multi-actor rehearsal with structural operations plus ledger for evidence-rich semantic episodes replacing brittle file/branch/commit records; storyboard serves as the Director's Planning Board using typed DAGs as first-class artifacts where dependencies, acceptance criteria, evidence requirements, tiered ranking, and status are all explicit; agentfederator acts as Casting Director with deliberate multi-LLM routing—frontier models for high-intent planning, quantized open models on capable hardware for execution velocity and cost. The central operating verb across every layer is ijbCRUD: provenance-aware and evidence-backed by construction where closure roots travel with artifacts.

What's Live Today

Real runs are already public. X replies and crossposts with multi-LLM consensus and evidence traces exist in the wild, along with a ledger distribution review governed by a living 22-unit DAG-TOML plan where the plan and its evidence became the shareable Exhibition record. sqry itself has been used in actual audits. Earlier articles and repo briefs on dag-toml and the production model are publicly available. These aren't vaporware or roadmap promises—they're real, usable artifacts you can point to.

What's Still Brewing

The fuller vision lives in internal verivusai-labs work: complete substrate (ijb, vault, ledger, integrity, meter and related crates), deeper studio refinements, and day-to-day use on larger efforts. The author notes this is "nights and weekends alongside the day job, completely disconnected" with learnings feeding one way only (#ihaveadayjob). The published verivus-oss artifacts are the current on-ramp, with pieces surfacing as they stabilize.

Key Takeaways

  • Collison's diagnosis of AI workflow friction is spot-on: semantic state survival and governance matter more than generation speed now
  • The "GNU Autotools × Notion" framing captures exactly what's missing from most current tooling
  • verivus-oss/vap provides a concrete open-source implementation mapping to every requirement Collison outlined
  • Real public artifacts exist—not just concepts—demonstrating multi-LLM consensus, DAG-TOML governance, and evidence-rich ledger episodes

The Bottom Line

Collison nailed the problem space. Someone's already building the answer in public with working code and real production runs behind it. If you've been feeling this friction—context that doesn't survive iteration, workflows collapsing into chat history, outputs you can't trust or provenance—you've got a landing zone to explore. The question now is whether the broader community engages with this direction or waits for Big Tech to ship something narrower.