One Week After Release — AI Operations Secretary Tasukiba's Phase 2/3 Roadmap

One week of operation made the picture sharper than I expected

It’s been a little over a week since the June 1, 2026 release.

What surprised me: the “next six months” I’d sketched vaguely before release came into focus after just one week of operation. Early-user behavior, the first feedback in, log patterns showing “fix this one thing and a lot gets easier” — when I wrote it all out, the roadmap stood up much more crisply than I’d expected.

The thing worth holding in mind, though, is the obvious one that gets missed: release isn’t “the end” — it’s “Phase 1 complete.” Phase 2 and Phase 3 are queued behind it.

Today opens Chapter L — post-release and roadmap. First topic: the Phase 2 / Phase 3 / beyond picture Tasukiba Knowledge Relay is drawing. I’m separating what I’m now confident I’ll do (after a week of operation) from what I’m honestly leaving open (“priority may shift”).

👉 Tasukiba product page

The overall map

The release plan has three phases.

Phase	Window	Main milestones
Phase 1 (MVP)	through 2026-06-01	Suggestion engine / chat semantic search / “Why?” feature (LLM re-ranking) / multi-tenant / billing foundation
Phase 2	release through 6 months	Stripe live billing / Dogfooding feedback / accuracy tuning
Phase 3	6 months through 2 years	Multilingual / public API

Capability expands step by step. All three signature features (suggestions, chat semantic search, “Why?”) shipped at release.

Phase 1 (MVP) — what’s live at release

What runs as of release:

Multi-tenant foundation (tenant creation / member management / system roles / project roles)
Project management (7-state one-way state machine / WBS / Gantt / task CSV import)
Asset management (knowledge / risks / issues / retrospectives / memos / attachments / visibility control)
Suggestion engine v1 (3-axis weighted score / Voyage embedding + pgvector / Graceful Degradation Mode)
Chat semantic search (pull past assets in natural language from the owl FAB at the bottom-right of every dashboard page)
“Why?” feature / LLM re-ranking (adds a “why is this related” explanation to suggestion candidates; Pro-only, Claude Sonnet)
Auth / security (NextAuth Credentials + MFA TOTP / tokenVersion multi-layer defense / security score 90/100 enforced)
Billing foundation (per-API-call usage billing / embedding usage billing / DB and file usage billing / ApiCallLog source of truth / Stripe Metered Billing wiring ready)

MVP-complete on the business-SaaS basics and all three signature features.

Phase 2 — within six months

1. Refining chat semantic search (the feature itself already shipped)

Phase 1’s suggestion engine took structured input (project purpose/background/scope). At release, we also shipped chat semantic search, which widens that input to natural-language chat (available from the owl FAB at the bottom-right of every dashboard page).

The experience:

User: “Wasn’t there a retrospective last month about resource shortage?”

Chat: “Here’s the matching retrospective:

2026-03-15: Sprint 5 retrospective

‘QA resources were short, test phase slipped’ Related risk: R-005 (resource estimation accuracy) Related knowledge: K-012 (criteria for introducing QA automation)”

Natural-language queries surface past assets — extending the suggestion engine in B-2 from structured forms to dialogue, and it’s already live. See the chat semantic search deep-dive for details. In Phase 2, the work here is search-accuracy tuning and UI refinement.

2. Stripe live billing

Phase 1 built the Stripe wiring, but production runs on manual invoice operations (risk hedge).

Phase 2:

Acquire Stripe production account
Identity verification, bank account registration
Webhook connection (live)
Migrate initial tenants to automatic billing

— reach “fully operational commercial SaaS.”

3. Dogfooding feedback

Post-release, I’ll be using Tasukiba for my own project management. Feedback from that goes back into Phase 2:

Suggestion engine precision tuning
UI improvements
Bug fixes
Feature request evaluation

This is the largest fuel source for Phase 2.

Phase 3 — six months through two years

1. Multilingual

Phase 1 is Japanese only. Phase 3 puts English support on the table.

UI translation
Voyage embeddings are already multilingual (mixed JA/EN OK in the same model)
Re-evaluate suggestion accuracy under multilingual conditions

2. Public API

Open Tasukiba’s API for external system integration.

REST API (OpenAPI schema)
Auth: API Key + OAuth2
Use cases:
- Call suggestions from an existing PM tool
- Hit Tasukiba from Slack / Teams chat
- Custom integrations (BPM / ERP)

Public API widens Tasukiba’s ecosystem.

Beyond Phase 3

Past three phases, with no fixed dates:

AI agent (operate Tasukiba via API as an agent)
Industry-specific templates (construction, manufacturing, SaaS dev)
Automated CSAT / NPS integration
Auto-categorization of knowledge + auto-generated reports

These get prioritized by Phase 1–3 progress and user feedback.

Why I publish the roadmap

Tasukiba publishes the roadmap.

Three reasons:

Users can predict “when does this feature show up?”
Users can decide “if I need this feature, I should pick another service”
Transparency builds trust

A lot of SaaS treats the roadmap as confidential. Tasukiba walks the other way. Same root as “the UI doesn’t lie” in B-4.

”The roadmap isn’t a promise” — stated

That said, the roadmap isn’t a promise.

User feedback shifts priorities
Technical discoveries surface new features
Market changes force direction changes

The roadmap updates periodically, with this in mind.

Users get the explicit label: “the roadmap is a plan, not a commitment.” As A-3 said about ambiguity by domain, future plans are a domain where ambiguity functions as breathing room.

First 30 days checklist

The first 30 days post-release matter.

Day 0: release day. Final production verification, Discord announcement
Day 1–3: initial-user signup follow-up, support
Day 4–7: usage-log analysis, catch unexpected behavior
Day 8–14: ship the first feedback fixes
Day 15–30: lock down the Phase 2 plan

That cadence sets up the post-release ops shape.

Summary

Phase	Window	Content
Phase 1 (MVP)	through 2026-06-01	Core capabilities + 3 signature features (suggestions / chat semantic search / “Why?”)
Phase 2	through 6 months	Stripe live / Dogfooding / accuracy tuning
Phase 3	through 2 years	Multilingual / public API
Beyond	TBD	AI agent / industry templates / ecosystem

Stepwise phases mean value ships at every phase.

Tomorrow: Dogfooding plan — using my own service on my own projects. The center of Tasukiba’s strategy: “don’t sell what you don’t use yourself.”

Two-year monetization check, but the service doesn’t stop — series part 14, Chapter K close
The suggestion feature — putting “the past you forgot” back on screen — the suggestion engine that became the foundation for all three signature features
Return autonomy to the user — a UI with no gatekeeper — root of publishing the roadmap

About Tasukiba

Tasukiba Knowledge Relay keeps expanding after Phase 1. See the product page for the current feature set and roadmap.