Research Roadmap

_Open investigations and blind spots. Updated 2026-07-13 (Commenting UX wishlist update — R4 expanded with concrete sub-questions; new blind spot #10)._

Active Investigations

R1: Team State Machine Adoption Patterns

Status: Hot — directly from state-machines-are-great pain point
Question: What patterns, arguments, and tools actually convince teams to adopt state machines?
Evidence: State machines are great, but hard to establish, Why aren't we using state machines more
Next step: Interview Eric and Alex (the skeptics) about what would convince them.

R2: The Mixed Logic Anti-Pattern Catalog

Status: Active — patterns identified, needs codebase mining
Question: What are the common forms of "mixed logic" (some logic inside state machine, some outside) and how do you migrate each one?
Evidence: State machines are great, but hard to establish block XjOJlPXm
Hypothesized patterns: useEffect Leak, Callback Escape, Middleware Creep, Event Ghost, State Split.
Next step: Mine the Seed codebase for real instances of each pattern.

R3: Agent State Machine Invariant Validation

Status: Active — spec written, needs implementation
Question: Can we validate an LLM agent session against the 25 formal invariants (I1–I25)?
Evidence: Agents and State Machines
Next step: Build checkInvariantI6 and apply to production KM agent logs.

R4: Comment Threading State Machine Design

Status: Active — updated with concrete sub-questions from the commenting UX wishlist (2026-07-13)
Question: What are the formal states of a commenting/threading interaction?
Evidence: What I want for the commenting experience — 5 concrete UX requests that map to state machine requirements

Concrete sub-questions from the UX wishlist:

Full-window discussions panel (request #1): What are the states of the panel? Open minimized / open full-window / collapsed? How does full-window mode affect comment rendering? Does it need a separate viewport state?

Block-addressable comments (request #2): How does the state machine handle comments that are themselves blocks? Can a comment block be in "draft" / "published" / "resolved" state independently of its parent document block? What does "block range" mean for a comment — is it a span over the comment's own block tree?

Mobile comment drawer (request #3): What are the touch gesture states? Drawer closed / partially open / fully open / pinned? How does scroll position interact with comment focus?

Mobile editor responsiveness (request #4): The editor is "laggy" for links, embeds, mentions. Is this a rendering state issue (e.g., re-render cascades on mobile viewport) or a composition state issue (e.g., embed resolution blocks UI thread)? Map the editor lifecycle as a state machine to identify where jank enters.

No-redirect commenting (request #5): The current architecture requires redirecting to a separate page for web comments. This is a routing state constraint. What would a SPA-style commenting overlay look like as a state machine? Compare with Remodeling the web thesis.

Next step: Map each of the 5 UX flows as machine states. Start with the block-addressable comment machine (request #2) — it's the most novel and connects to Signature Idea #3.

R7: Practical-to-Theoretical Bridge for Agent State Machines

Status: Active — cluster connected, bridge content missing
Question: What is the most effective narrative to bridge from Cursor background agents to formal state machine theory?
Evidence: The cluster is fully connected. How to Background Agents links to formal theory, architecture, hub, and product roadmap. The practical-to-theoretical bridge is the next gap.
Next step: Prototype a single-file React + XState v5 example modeling a simple background agent lifecycle.

R8: Practical-to-Product Bridge — Cursor Agents to Seed Agents

Future Investigations

R5: Visual State Machine Debugging for Non-Engineers

Priority: High — addresses the team-convincing problem. The mixed logic discovery (R2) adds urgency.

R6: State Machines vs. LangGraph/Mastra/Vercel AI SDK

Priority: Medium — needed for the definitive guide.

Closed Investigations

_None yet._

Blind Spots

No community feedback on agent determinism idea — The trigger docs ask readers for opinions but have no public comments yet

No comparison with existing agent frameworks — LangGraph, Mastra, Vercel AI SDK

No concrete testing patterns for state machines — How do you unit test an agent state machine?

No adoption metrics — Has any team successfully adopted state machines after reading these docs?

No costing analysis — What does running an agent state machine cost vs. traditional approaches?

No formal definition of "mixed logic" — Needs crisp definition and detectable characteristics

No teachable tutorial between practical and theoretical poles — The cluster is fully connected but inaccessible to newcomers

No mapping between Cursor agent concepts and Seed agent architecture — Cross-link exists, no explanatory content

Status: New — revealed by "Vibe code like a PRO" update
Question: What is the most effective framing to connect the "vibe coding" trend (mainstream, huge search volume) to the deterministic agent state machine thesis without diluting technical depth?
Evidence: Vibe code like a PRO title explicitly uses "Vibe code." The ai-dev-tasks repo has 7.8k stars. The term "vibe coding" (coined by Andrej Karpathy) has become a category. No serious piece has argued that vibe coding's future is deterministic agent pipelines with state machines.

Key questions:

What is the minimum technical depth for a "vibe coding" audience to grasp the state machine argument?

What existing vibe coding content exists and what framing gaps are open?

Can "From Vibe Code to Deterministic Agents" serve as a lead magnet for the manifesto?

Next step: Search for existing "vibe coding + determinism" or "vibe coding + state machines" content. If none exists, the framing piece is unopposed.

No "vibe coding" framing bridge exists — The mainstream vibe coding audience has no connection to the deterministic agent thesis. (2026-07-13)

No content comparing Seed's comment model with Slite/Coda/Linear/Google Docs — The competitive research corpus exists but is unsynthesized. A comparison piece would position Seed's block-level comment approach against established patterns. (NEW — 2026-07-13)

---

_Last updated: 2026-07-13 (Commenting UX wishlist update — R4 expanded). Maintained by: Strategist Agent_