The "Interface Notions" document explicitly calls out the goal of making AI agents stick to codebase conventions. This is the highest-leverage artifact we can produce.

Impact: High — solves the "generating random code that nobody will be able to maintain" problem head-on

Effort: Medium — the catalog exists (100+ components inventoried), needs translation into agent instruction format

Confidence: High — directly requested by the document author

Why it matters: As more of Seed's code is written with AI assistance, having a formalized conventions guide prevents quality decay at scale. This compounds: each agent generation uses better conventions.

Dependencies: Interface Notions doc is stable; needs agreement on agent instruction format

Next action: Extract the component hierarchy into structured agent prompt templates

Supporting evidence:

Interface Notions — "I would be more comfortable if I learned the fundamentals and concepts of the interface. This way, I make sure the Agent Sticks to our conventions"

Projects page — Interface Notions now embedded as a project artifact

---

The document machine is already a 465-line XState v5 implementation used across desktop and web. Extracting it as a standalone package would serve as both a reference implementation and a marketing asset.

Impact: High — establishes Seed's engineering credibility in the XState/state-machines ecosystem

Effort: Medium-High — extraction, testing, documentation, npm publishing

Confidence: High — the machine exists, is documented, and is referenced across multiple documents

Why it matters: Compounds with the State Machines content series. A real, open-source state machine from a production app is the best possible advertisement for the technique.

Dependencies: Need a clean boundary between app-specific and generic logic

Next action: Audit the machine for app-specific dependencies; draft the public API surface

Supporting evidence:

Notes about how the document machine should work — detailed state lifecycle spec

Project Plan — New Publish mental model — implements the machine

---

The Graduated Trust proposal demonstrates Seed's platform capabilities (contacts, capabilities, documents) solving a real industry crisis. It should be repackaged as a platform case study.

Impact: Medium-High — positions Seed as infrastructure for trust, not just publishing

Effort: Low — the content exists as a long-form proposal; needs editing and restructuring for a different audience

Confidence: High — the mapping between proposal needs and Seed primitives is explicit and well-argued

Why it matters: This is Seed's most ambitious platform narrative. It shows what Seed enables that no other system can.

Dependencies: None for the case study; implementation would be larger

Next action: Extract the "Seed primitives" section into a standalone "How Seed enables Web of Trust" piece

Supporting evidence:

Graduated Trust — full proposal with explicit Seed connections

---

The editor block rendering fix is well-documented but buried inside a tech talk. The content-start alignment principle deserves standalone treatment as a design principles document.

Impact: Medium — evergreen reference for the editor team and contributors

Effort: Low — content exists, needs extraction and visual polish

Confidence: High — the principle is already articulated and the fixes are in production

Why it matters: A clear, visual design principles doc prevents future regressions and serves as onboarding material for new editor contributors.

Dependencies: None

Next action: Extract the principle and before/after visuals into a standalone doc at /design/content-start-alignment

Supporting evidence:

Improving Editor Block Rendering — full analysis embedded in the projects page

---

The UI with State Machines hub has 4+ articles, case studies, integrations, and patterns. The next growth vector is bundling these into a structured learning path or video course.

Impact: High — compounds existing content into a higher-value asset

Effort: Medium — requires curriculum design, video production, and a landing page

Confidence: Medium-High — audience exists, content is proven; uncertainty is in format and distribution

Why it matters: A course converts casual readers into practitioners. Practitioners become advocates for the state machine approach, which reflects well on Seed.

Dependencies: Newsletter list growth, video equipment/editing

Next action: Audit existing articles and identify gaps for a complete curriculum

Supporting evidence:

UI with State Machines hub — full site with content, patterns, integrations, case studies

---

Embed feature demo — too specific, better as part of broader documentation

Electron tRPC analysis — valuable but narrow; better as a one-off post than a strategic asset