Infrastructure before prototype
16-phase agent migration, event bus, orchestrator, gateway API — all built before a single user onboarding flow shipped.
Desk-flip autopsy
What killed v2.
26 apps archived, zero active, 1,647 commits across 16 phases. The Poketo monorepo accumulated extensive infrastructure but never shipped a user-facing product. This is the autopsy.
What Went Wrong
ICP scope creep
Onboarding expanded from 1 ICP path to 5 without proving the first one worked with real users.
Planning > shipping
Doc-reframe commits outnumbered feature commits. Planning-to-implementation ratio grew past 1:1.
External blockers drifted
Upstash Redis provisioning blocked Phase 8 for 36+ days with no fallback or descoping.
"Shipped" wasn't usable
13 specs marked "Shipped" meant backend contracts exist — not that a user could experience the feature.
By the Numbers
Apps archived
Work, Report, Chat, Infinite, Brain, Focus, Matrix, Timeline, Space, and 17 more — all moved to archive/v2/apps/.
Spec files written
13 marked "Shipped" based on backend contract completion, not user-accessible features.
Non-feature commits
Docs, chore, and test commits with no corresponding feature implementation. ~200 were pure scope-reframe commits.
Salvageable Assets
16 valid research & spec documents carry forward. 4 partially valid. 3 stale.
| Category | Documents | Status | Key content |
|---|---|---|---|
| Market research | ICP, competitive analysis, positioning, GTM, monetization, assumptions | Valid | Agency ICP (5-20 employees), 29-competitor landscape, April Dunford positioning, PLG strategy, pricing tiers |
| Concept briefs | Poketo Core, Work | Valid | Modular suite with animal mascots, Work as PM module (boards, lists, cards, agent sidebar) |
| Spec domain models | CRM, quote-to-cash, onboarding, share links, agent modes | Valid | Pipeline CRM CRUD, billing scope boundaries, ICP-scoped onboarding loop, client portal |
| Brand | Strategy, notes | Valid | Mascot duo (Poke/Poket), feature naming, app integration |
| Metrics & personas | North Star metric, Agency Ops agent spec, Core ICP | Partial | Concepts sound but depend on product shape or are untested against prototype |
| Implementation artifacts | STATUS.md, orchestrator, event tracking | Stale | Document v2 code state, not reusable domain knowledge |
Lessons for the Fresh Start
Ship a prototype before writing specs
v2 wrote 37 specs and built backend contracts for 13 "shipped" features with no user-facing app consuming them. Prove one user journey works end-to-end before expanding scope.
One ICP, one app, one aha moment
v2 planned five ICP onboarding paths before proving one. Pick the strongest ICP (agency founder, 5-20 employees) and build exactly one app (Work) to one aha moment.
Cap planning overhead
When doc-reframe commits outnumber feature commits, the project optimizes for legibility over progress. Limit planning to what's needed for the next 2 weeks.
Resolve or descope blockers in 3 days
The Upstash Redis blocker sat for 36 days. If an external dependency blocks for >3 days, ship with the fallback and revisit later.
"Shipped" means a user can experience it
Backend contracts, passing tests, and tRPC routes are not shipped features. Definition of done must include a working UI a real person can use.
No shared infra before the first consumer
Event bus, orchestrator, gateway API, agent tooling — all built before the first app needed them. Build these only when the second app creates genuine duplication.
v3 Bootstrap Path
What to carry forward
- Research artifacts (ICP, competitive analysis, positioning, journey map, GTM, monetization, metrics, assumptions)
- Concept briefs (Poketo Core, Work)
- Brand strategy & mascot system
- Select spec domain models (CRM, billing, onboarding, share links, agent modes)
Do not carry forward
- Code, dependencies, or database schemas
- Infrastructure config or gateway setup
- Implementation-tied specs (orchestrator, event tracking, STATUS)
- The 26-app architecture — start with one
Recommended Build Sequence
Revalidate ICP
Rerun ICP analysis against v3 scope. One segment: agency founder, 5-20 employees.
Work prototype
Standalone Next.js app with basic kanban UI. No shared infrastructure, no agent layer.
Onboarding-to-aha
Prove one journey: configure → seed → triage → action with a single ICP path.
Agent capabilities
Add AI assistance only after the manual workflow works and has real usage.
Second module
Expand only after Work has real users. Likely CRM or client portal.
Source Artifacts
Reference documents that informed this autopsy.